Full text of "An Introduction To Behavioran Introduction To Behavior Theory Theory"

See other formats

Author

Issue

Date

Borrower’s

No.

Borrower s

f: fr

A- • ; 'V-,

'■ffe-i v, ..

‘i

\*&gk •

* • .

-z&f'
• *- ■ .*.

:■

Principles of Behavior

AN INTRODUCTION TO BEHAVIOR THEORY

CLARK L. HULL

—

Late, of Yale University

APPLETON-CENTURY-CROFTS

Division of Meredith Publishing Company

NEW YORK

D. APPLETON-CENTURY COMPANY, INC.

All rights reserved. This hook , or parts
thereof , must not be reproduced in any
form without permission of the publisher.

628 -8

Library of Congress Catalogue: 66—28615

•E-47202

PRINTED IN THE UNITED STATES OF AMERICA

PREFACE

As suggested by the title, this book attempts to present in an
objective, systematic manner the primary, or fundamental, molar
principles of behavior. It has been written on the assumption that
all behavior, individual and social, moral and immoral, normal and
psychopathic, is generated from the same primary laws; that the
differences in the objective behavioral manifestations are due to
the differing conditions under which habits are set up and function.
Consequently the present work may be regarded as a general intro-
duction to the theory of all the behavioral (social) sciences.

In an effort to insure its intelligibility to all educated readers,
the complicated equations and other more technical considerations
have been relegated to terminal notes, where they may be found
by the technically trained who care to consult them. The formal
verbal statements of the primary principles are presented in special
type at the ends of the chapters in which they emerge from the
analysis. A convenient glossary of the various symbols employed
is provided.

There remains the pleasant duty of recording my numerous
obligations. First in order of importance is my gratitude to the
Institute of Human Relations and to its Director, Mark A. May,
for the leisure and the generous provision of innumerable accessory
facilities which have made possible the preparation of this work,
and for the stimulation and instruction received during several
years of Monday night Institute staff meetings. Contributing to
the same end have been the stimulation, criticism, and suggestions
given by the members of my seminar in psychology in the Yale
Graduate School. Few things have been so pleasant and so profit-
able scientifically as the contacts with the brilliant and vigorous
young personalities encountered in these situations. The most of
whatever is novel in the pages of this book has been generated in
one way or another from these contacts.

To certain individuals a more specific debt of gratitude must
be acknowledged. All the original figures were drawn by Joy
Richardson and Don Olson. Bengt Carlson fitted equations to the

PREFACE

numerous empirical curves, and is responsible for all of the more
technical mathematical material which appears in the terminal
notes. Marvin J. Herbert prepared the subject index. To my
laboratory research assistants of the last ten years, Walter C.
Shipley, St. Clair A. Switzer, Milton J. Bass, Eliot H. Rodnick,
Carl Iver Hovland, Douglas G. Ellson, Richard Bugelski, Glen L.
Heathers, Peter Arakelian, Chester J. Hill, John L. Finan, Stanley
B. Williams, C. Theodore Perin, Richard O. Rouse, Charles B.
Woodbury, and Ruth Hays, I am indebted for the conscientious per-
formance of numerous experiments which were especially planned
for this work, and which naturally make up a considerable propor-
tion of the empirical material employed. The loyal cooperation and
kindly day by day suggestions and criticisms of these splendid

young people have made that phase of the task a rarely satisfying
labor.

Another and smaller group of individuals have given invalu-
able aid in the preparation of the manuscript. Eleanor Jack
Gibson read much of an early version of the manuscript and made
helpful suggestions. Irvm L. Child read the manuscript in a late
stage of revision and made numerous valuable criticisms and sug-
gestions. To Kenneth L. Spence I owe a debt of gratitude which
cannot adequately be indicated in this place; from the time when
the ideas here put forward were in the process of incubation in
my graduate seminar and later when the present work was being
planned, on through its many revisions, Dr. Spence has contributed
generously and effectively with suggestions and criticisms, large
numbers of which have been utilized without indication of their
origin. Finally, to Ruth Hays I am deeply indebted for the tran-
scription of hundreds of pages of unbelievably illegible handwrit-
ing, for the preparation of the name index, and for absolutely
indispensable assistance with the formal aspects of the manuscript.

„ C. L. H.

New Haven

FOREWORD TO THE
SEVENTH PRINTING

In the short span of 11 years there appeared in the Century
Psychology Series three books that not only decisively determined
the course that one area of psychology was to take over the ensuing
20 years, but also literally instigated most of the research carried
out in it. The books, in order of their appearance, were Tolman’s
Purposive Behavior in Animals and Men (1932), Skinner’s The Be-
havior of Organisms (1938) and Hull’s Principles of Behavior
(1943). The area of psychological research with which they were
primarily concerned was the behavior of animals in learning situa-
tions, a set of phenomena, which, as the earlier writings of Thorn-
dike and Watson attest, has been by tradition the favorite starting
place of American psychologists’ attempts to formulate a systematic
theoretical framework. Not only is this custom carried on in these
books, but all three may also be said to have continued the same
objective behavioral approach that characterized their two illustri-
ous predecessors.

In considering what comments to make on Hull’s contribution
to this series, it seemed appropriate to begin by discussing more
generally all three books. Especially desirable, it seemed, was an
overview that would not only clarify their relation to the earlier
behaviorism of Watson but would also examine in what respects the
three systems were similar or different. To imply that all three books
are behavioristic in outlook does not, of course, either identify their
authors as behaviorists in the Watsonian pattern, or as having de-
veloped more or less identical positions. The manner in which
these formulations differed from the behaviorism of Watson is well
known as the distinction between molecular and molar behaviorism,
of which more later. That these three new versions of behaviorism
were strikingly different from each other in certain features is abun-
dantly clear from the fact that their authors were often found to be
on opposite sides of some of the sharpest and most highly publicized
controversies in the area of learning theory. Examples that come
most readily to mind are such issues as cognitive organizations

(Tolman) versus S-R association (Hull), S-S contiguity (Tolman)

• •

Vll

Vlll

FOREWORD TO THE SEVENTH PRINTING

versus S-R reinforcement (Hull-Skinner), systems that employed
theoretical constructs such as intervening variables ( Hull-Tolman)
versus the radical empiricism of Skinner which is supposed to have
shunned all theorizing. Unfortunately, the overemphasis on these
and other issues, some of which developed after the publication of
the books, has lessened recognition of the extent to which the three
formulations are basically similar in their systematic approach. As
an antidote to this overemphasis of the differences in some of the
details of these systems, the following analysis attempts to center
attention on the common framework in which they were formu-
lated and the rather remarkable similarities in the methodological
principles that guided them.

Thoroughly behavioristic in approach, the focus of interest of
all three books is in behavior per se and not in the mind or con-
sciousness. In this regard, all three were, of course, following Wat-
son. However, in sharp contrast to the molecular, quasi-physio-
logical conception developed by Watson, the neo-behaviorisms of
these books offered a molar approach that not only defined re-
sponses in a different fashion, but also shifted interest away from
the neurophysiological determinants of behavior to events happen-
ing in the environment. Thus, in place of Watson’s narrow defini-
tion of behavior in terms of detailed movements or patterns of
muscular contraction, a new type of specification in terms of the
achievements or outcomes of the organism’s acts was employed.
Hull and Skinner differed slightly from Tolman in this matter in
that their repsonse definitions were limited to the specification of
effects on or with respect to the environment immediately present
at the time of the response. On the other hand, Tolman insisted, in
principle, on going beyond such immediate effects and employing
subsequent environmental events, such as the attainment of later
goal objects, in specifying his behaviors. However, in actual prac-
tice his response definitions, especially in the case of his animal
experiments, differed little from those of Hull and Skinner.

With this common focus of interest on behavior in relation to
environmental events, it is perhaps not surprising to find that all
three also had similar conceptions as to the task confronting them.
Indeed they employed essentially the same kind of formula to
represent the end product they sought. In terms of the symbols
used by Skinner, this was the now familiar equation: R = f (S, A).
In this general expression, representing the kinds of laws sought in
learning experiments, R stands for the dependent response measures.

FOREWORD TO THE SEVENTH PRINTING

while S and A have reference to two kinds or classes of environ-
mental events that constitute the independent variables. Thus S
includes the particular patterns of stimulus events ( objects, etc. ) in
the physical or social environment present at the moment of the
organisms response, while A has reference to two subclasses of
environmental happenings. The first of these subclasses is comprised
of stimulus events that are the outcomes of the organism’s re-
sponses. Known as incentive variables, these include such things as
rewards of varying magnitude, nonrewards and punishments. In
effect, such stimulus outcomes of responses constitute antecedant
variables that determine, in part, the properties of subsequent be-
havior in the situation. The other subclass contains certain kinds of
environmental events that typically occur outside the experimental
situation itself. Examples are the amounts of handling an animal
subject has experienced, the schedule and amounts of food and
water feedings prior to and during the conduct of the experiment,
the time elapsing between successive trials, the administration of
noxious stimuli prior to the experiments, and so on. These particular
kinds of environmental manipulations are known as motivational
variables.

In terms of this common formula, all three of these psychologists
conceived their task to be that of discovering and formulating the
many laws relating these different classes of experimental variables
to behavior. In the initial stages an important aspect of this en-
deavour is the defining of potentially significant behavioral measures
and the identification and specification of the relevant environ-
mental variables. The early researches of Tolman and Skinner were
primarily concerned with such matters (e.g., Skinners rate measure).
By the time of Hull s book, the stage had been reached in which
there was greater emphasis on the determination of precise quanti-
tative laws. These latter involved both simple laws relating an R
variable to single environmental variables, or complex laws de-
scribing the joining action of several variables on behavior.

But the resemblances in the approaches taken in these three
books do not end here. Just as the classes of experimental variables
and kinds of empirical laws proposed for investigation were similar,
so were, in principle at least, their conceptions as to the role of
theory and the form it should take in the total scientific enterprise.
As far as Hull and Tolman are concerned this statement would
probably meet with fairly general assent. Not only did both at-
tempt to make considerable use of theory but in his Principles Hull

FOREWORD TO THE SEVENTH PRINTING

acknowledged that his theoretical constructs were of the type that
Tolman has been credited with introducing into psychology and to
which he gave the label “intervening variable.” To include Skinner
as a party to this accord, however will no doubt come as a consider-
able surprise, if not shock, to some. Especially will this be the case
with those persons who have forgotten Skinner’s earlier writings or
whose reading has been limited to what has appeared since 1950.
Certainly since that date Skinner has become not only ultraposi-
tivistic, but increasingly antitheoretical.

However, a careful reading of Skinner’s earlier writings, includ-
ing The Behavior of Organisms, will show that he not only engaged
in a kind of theorizing, but that the type of theoretical construct he
used was the same as that which Tolman, in principle, advocated
and Hull, in practice, employed. As a matter of fact a strong case
can be made for the proposition that Skinner was the first to describe
the role of intervening variables in psychological theorizing (Skin-
ner, 1931) although he did not call them such. Indeed, his treat-
ment in this earlier article and in his 1938 book of the nature and
function served by such “hypothetical middle terms,” as he referred
to them, is the most lucid and intelligible the writer has ever seen.

While space limitations will not permit an extensive discussion
of Skinner’s treatment of these theoretical constructs, some exposi-
tion of the manner in which he conceived them will serve to show
their similarity to the intervening variables of Tolman and Hull.
This kind of construct was introduced by Skinner in connection with
what he called dynamic laws, those relating R variables to the class
of antecedent A variables, as distinguished from static laws, which
relate R variables to the class of S variables. They are particularly
useful, Skinner pointed out, when a number of different experi-
mental operations lead to the same kind of observable changes in
the values of different measures of a single reflex ( S-R relation ) or
in the concurrent changes in strength of a group of different reflexes.
The orderliness or lawfulness of these concurrent changes permit the
inference of a common effect or “state” underlying them. Examples
given by Skinner are such “hypothetical intermediate terms” or
“states” as reflex strength, which may be identified with Hull’s con-
struct of excitatory potential (E), hunger drive, emotional drive,
and reflex reserve, which also have their counterparts in both Hull’s
and Tolman’s systems. As Miller (1959, p. 277) has pointed out,
the gain in efficiency, measured in terms of minimizing the number
of relationships among the variables required to describe a realm of

FOREWORD TO THE SEVENTH PRINTING

data, becomes greater and greater as both the number of experi-
mental operations and the number of different empirical response
measures go beyond two. When one considers the many different
kinds of operations that can be employed to specify the different
drives and kinds of reward outcomes, as well as the number of dif-
ferent response measures in the many types of learning situations,
it is evident that this gain could be considerable. It was this feature
of a theory employing intervening variables that the writer ( 1957,
p. 87) had in mind in the statement “intervening variables of learn-
ing theory serve as general or abstract terms that are applicable to
a variety of situations.” This description would be more complete if
to it were added “and a wide variety of experimental operations.”

We thus have the interesting fact that each of the systems for-
mulated in these three books offered more or less the same concep-
tion as to the nature of the integrating theoretical concepts needed
at the time. Strange as it may seem, neither Hull nor Tolman ever
gave any evidence that they were aware of Skinners excellent anal-
ysis and use of these kinds of concepts in the organization of be-
havioral laws. One reason for this, undoubtedly, is that during this
period at least Skinner’s book did not receive the attention it de-
served. Whether as a result of this low level of reinforcement or
because of other factors, Skinner himself all but abandoned this
earlier interest in the search for the abstract kinds of laws that all
sciences have traditionally sought. Apparently believing that psy-
chology is not ready for such theorizing, he turned his efforts almost
entirely in the direction of what is described as the experimental
analysis and control of individual behavior. Essentially this involves
attempts to discover in each concrete situation what the relevant
conditions of behavior are and then to proceed to modify or shape
the behavior in the desired direction by using already available
knowledge concerning the effects of applying various kinds of rein-
forcing procedures. Since reinforcements play an exceedingly po-
tent role in modifying behavior, a technology based even on such
knowledge is bound to be strikingly effective, especially in com-
parison with what the traditional procedures in the area are likely to
have been. The use of teaching machines in education is a case in
point. Quite in contrast to his earlier experience with theoretical
concepts, this recent activity of Skinner has been on the receiving
end of an extremely high frequency schedule of reinforcements,
From the point of view of the writer this is most unfortunate as if
renders unlikely the probability of his interests ever shifting back

xii FOREWORD TO THE SEVENTH PRINTING

the task of developing the more abstract kind of knowledge that all
sciences strive to achieve, a task for which The Behavior of Or-
ganisms showed him to be brilliantly qualified.

Quite in contrast to Skinner, who, as has just been described,
soon rejected as permature even the level of theorizing represented
by intervening variables, Hull persisted in his theoretical efforts and
attempted to extend the formulation developed in the Principles to
more complex areas of behavior. An understanding of the role and
purpose of this aspect of Hull’s work as well as certain other of his
special interests— for instance his desire to provide for the possi-
bility that his molar psychological system could eventually be inte-
grated with the more molecular area of neurophysiology at one
level and the even more molar social sciences at the other— is
essential to an appreciation of what he tried to accomplish in the
present book. If one may judge from some of the reviews of it,
there was considerable misunderstanding of the meaning of the
proposed system. Some of these were quite bitter. Thus from the
Tolman camp came cries that despite Hull’s claim (p. 20) that his
theory was molar in nature, in reality it revealed a strong molecular
bias, which was somehow bad. In support of this assertion, they
cited a number of chapters (e.g., Ill, IV and V), the contents of
which are primarily physiological in nature. Also, attention was
called to the fact that some of postulates referred directly to neural
events, while even those that were defined mathematically on the
basis of behavioral laws often included additional, and to the critic,
superfluous physiological connotations (e.g.. Postulate 4). Skinner
(1944, p. 278) reacted in a similar manner when he wrote “The
exigencies of his [i.e., Hull’s] method have led him to abandon the
productive (and at least equally valid) formulation of behavior at
the molar level and to align himself with the semineurologists.”

It is a little difficult to understand why these critics did not see
what Hull was up to, for while he acknowledged that a theory of
behavior based on both molecular (neurological) and molar (be-
havioral) principles would be more satisfactory than one founded
upon molar considerations alone, he clearly recognized that our
present knowledge of neurophysiology would not permit such a
theory. However, as he took great pains to point out, molar and
molecular are relative terms, psychology being molar relative to
physiology but molecular relative to sociology. Hull firmly believed
that the day would arrive when a single comprehensive theory
would be developed that would encompass all of these areas. He

FOREWORD TO THE SEVENTH PRINTING

simply was eager to provide for the possibility that the molar
(behavioral) theory that he was developing would, in addition to
accounting for behavior in the area on which it was based, also
relate in one way or another to areas above and below in the
hierarchy of behavior sciences. As will be brought out later he
proposed to use the hvpothetico-deductive method as the means of

deriving more complex social behavior. The only means he had
available to indicate possible relations of his behavior theory to
neurophysiology was to suggest physiological processes that might
underlie his behaviorally specified intervening variables. However,
as the present writers theorizing has demonstrated, no psychologist
is under obligation to accept or even pay attention to these physio-
logical implications. Since their correctness or incorrectness has no

bearing on the adequacy of the theory as a behavioral theory, one
should be able to take them or leave them. The reaction of the

critics reflected the degree of their own antiphvsiological biases.

Turning now to Hull’s molar theory, it was developed in much
the same fashion as Skinner’s except that it involved as a base the
much wider range of behavioral laws provided by different kinds of
experimental situations in the areas of both classical and instru-
mental conditions. Since the number of variables involved, par-
ticularly the number of different response measures, was much
larger than in Skinner’s case, the abstract logical constructs intro-
duced by EIull allowed for an even greater gain in efficiencv of the
kind described earlier. Such a theory consists of a chain of interven-
ing variables interpolated between the independent environmental
variables and the dependent response variables. It involves the
assumption of a multiplicity of functional relationships among the
theoretical constructs and between them and the independent and
dependent experimental variables.

In its initial formulation the theory is, of course, a purely ad hoc
affair. Nevertheless, it is possible to test it even within the experi-
mental settings in which it was developed, for the implications of
the theory are always a joint function of the theoretical assumptions
and the particular initial or boundary conditions of each experiment.
By varying these boundary conditions, new implications (theorems
as Hull called them) may be derived that can then be tested by
further experiments. Hull’s book contains a number of illustrations
of such tests of his theory.

Much more interesting and significant, however, is the extension
of the theoiy to new' phenomena, to behavior in different situations

XIV

FOREWORD TO THE SEVENTH PRINTING

from those in which it was developed. In some instances the new
situation merely involves novel combinations of the boundary con-
ditions, in which case the derivation is straight forward. More
complex situations, however, frequently require the formulation of
what are known as composition rules. An example of such a rule
is Hull’s 16th postulate, which states how the excitatory strengths of
two incompatible S-R’s interact with one another. If additional
experimental variables are introduced, new intervening variables
with their relationships to the old must be guessed at. In this event,
the development of a more comprehensive theory typically involves
a series of approximations with different assumptions being succes-
sively tried out.

This description of the kind of theory Hull attempted to develop
and the manner in which it is extended to areas of behavior other
than that on which it was originally formulated has been presented
in order to make clear how the Principles fitted into his overall
plans. In this book, only the first step was undertaken— the develop-
ment of a theoretical structure that was able to derive and inter-
relate the available laws in two simple types of learning, classical
and instrumental conditioning. Except for some illustrative exam-
ples, e.g., the derivation of some phenomena in connection with
learning in a choice situation involving different delays of reinforce-
ment, Hull did not attempt to extend the theory beyond its original
base.

Before the completion of the Principles , however, Hull was al-
ready at work on the second volume of a series of three books that
he hoped to write. Unfortunately, poor health slowed his progress
towards this goal and he was able only to complete the second
volume, A Behavior System , before his death. This second book
not only presented a revision of his earlier theory, but extended it
to the derivation of more complex kinds of individual ( nonsocial )
behavior. Included among these phenomena were various kinds of
simple and compound trial and error learning, different types of
discrimination learning, maze learning, acquisition of skills, and even
such little-investigated matters as value and valuation, language as
pure stimulus acts, and reasoning. The third volume was to have
presented a parallel treatment of the more elementary forms of so-
cial and group behavior.

A discussion of the methodology of Hull’s theorizing would not
be complete without some comment on his use of the postulate or.

FOREWORD TO THE SEVENTH PRINTING

as it is sometime called, hypothetico-deductive method. His long-
time interest in the kind of theory that explained phenomena by
deriving them from explicit assumptions by means of rigorous logi-
cal procedures is evident in one of his earliest papers on learning
(1930). Subtitled “A study in psychological theory,” this article
questioned the vagueness of existing psychological theories, par-
ticularly those of the Gestalt psychologists, and offered as a chal-
lenge a theory of simple trial and error learning. The style of
presentation in this article did not involve stating the basic assump-
tions as formal postulates, but rather used the narrative form. Fol-
lowing the reading of an article by Einstein in the initial volume
of the journal. Philosophy of Science, Hull became an enthusiastic
proponent of the formal postulate method. He carefully studied
Newton’s Principia and endeavored to use it as his model of what a
psychological theory should be. This influence is revealed in the
series of so-called miniature postulate systems, culminating in the
monograph entitled, Mathematical-Deductive Theory of Rote
Learning, that appeared in the period 1935-1940.

Unfortunately, a certain amount of confusion developed as to
whether these postulate systems represented a different type of the-
ory from that which employed intervening variables. The fact that
Hull made no mention of intervening variables in any of these
formulations added to the impression that it did differ. Further-
more, the occurrence of terms like undefined concepts and defini-
tions, which defined new terms, led to the comparison of Hull’s
postulate systems with the type of formal language systems originally
developed in logic and mathematics (e.g., Hilbert’s Axiomatics).
By interpreting the purely formal terms of such systems by means
of so-called coordinating definitions to the concepts of an empirical
system of laws, scientists in advanced fields of physics had found
this type of theory to be a powerful technique for the integration
of previous unrelated areas of knowledge. However, an analysis of
Hull’s postulate systems by Bergman and the writer (Bergman and
Spence, 1941) showed them not to be of this character. They did
not begin with a set of purely formal (i.e., undefined) terms that
have no other meaning than that provided by a set of implicit
definitions (axioms) from which are then derived new terms and
theorems made testable by means of coordinating definitions.
Rather, Hull began with operationally defined experimental vari-
ables in terms of which he introduced his theoretical constructs.

XVI

FOREWORD TO THE SEVENTH PRINTING

Thus an examination of the postulates in the Principles will show
that they are statements of the functional relationships linking the
intervening variables to the experimental variables or to each other.

Following the article by Bergman and the writer, Hull explicitly
acknowledged that his constructs were of the intervening variable
variety. His use of the postulate method should be understood,
then, as merely a means of making more explicit the assumptions
made by the theorist and assuring as rigorous and complete as
possible the deductive elaboration of their consequences. In his
Principles Hull compromised with the formidable style of his earlier
postulate systems and presented his theory in an informal manner.
Verbal statements of the postulates were given at the end of a chap-
ter and their mathematical formulation was relegated to terminal
notes. In order to keep the book within reasonable bounds, only
representative implications of the theory for classical and instru-
mental conditioning were presented, while those for more complex
areas were reserved, as described above, for subsequent volumes.

In bringing these comments to a close one further and most
important characteristic of Hull's theorizing should be emphasized,
namely its provisional character. Hull understood well both the
tentative and incomplete nature of the empirical foundations on
which he had dared to go ahead and also the very great likelihood
that the principles he put forward would be found to be defective
in one respect or another. He was not afraid of being wrong, for
he knew that science is to some extent a trial and error process in
which errors as well as successes will occur. He deplored vagueness
and attempted to make his own theory as precise and clear as possi-
ble, partly as a means of detecting errors quickly and thus hastening
a correct formulation.

Finally, no one knew better than he the enormity of the task he
was undertaking or appreciated more the limited progress that
would be made in his lifetime. His hope was that the younger
generation of psychologists would be stimulated by this beginning
to engage in the kind of research and theory necessary to attain the
goal he sought, a behavioral science at the level of development
achieved by the physical sciences in the age of Galileo and Newton.
That this book greatly stimulated and influenced the direction of
research in the 10 years that he lived following its appearance is
shown by the remarkable statistic that during this period approxi-
mately 70 per cent of the studies reported in the Journal of Experi-

FOREWORD TO THE SEVENTH PRINTING xvii

mental Psychology and Journal of Comparative Psychology made
reference to his writings. The large majority of these were to the
Principles of Behavior.

Kenneth W. Spence

REFERENCES

Bercman, G., & Spence, K. W. “Operationism and theory in psychol-
ogy.” Psych. Rev., 1941, 48, 1-14.

Einstein, Albe.rt. “On the method of theoretical physics.” Philos, of
Science, 1934, 1, 163-169.

Hull, L. L. “Simple trial-and-error learning: A study of psychological
theory.” Psychol. Rev., 1930, 37, 241-256.

Miller, N. E. “Liberalization of basic S-R concepts: Extensions to con-
flict behavior, motivation and social learning.” Psychology: A Study
of Science, Study I, Volume 2, S. Koch (Ed.) McGraw-Hill Co.,
New York, 1959.

Skinner, B. F. “The concept of the reflex in the description of behavior.”
J. General Psychol., 1931, 5, 427-458.

Skinner, B. F. “Review of Hull’s Principles of Behavior ." Amer. J.
Psychol., 57, 276-281.

Spence, K. W. “The empirical basis and theoretical structure of psy-
chology.” Philos, of Science, 1957, 24, 97-108.

Title

Author.

Accession No

r-«n kt-

Issue

Date

ACKNOWLEDGMENTS

For permission to quote, the author makes grateful acknowledg-
ment to the following publishers: A. and C. Black, London; Clark
University Press; D. Appleton-Century Company, Inc.; Harcourt,
Brace and Company ; Liveright Publishing Corporation ; Macmillan
Company; National Society for the Study of Education; W. W.
Norton and Company; Oxford University Press; Williams and
Wilkins Company; Yale University Press.

The author also acknowledges with thanks permission to quote
and reproduce figures from the following journals: American Jour-
nal of Physiology ; Archives of Psychology; British Journal of
Psychology; California Publications in Psychology; Journal of
Comparative Psychology; Journal of Experimental Psychology;
Journal of General Psychology; Journal of Genetic Psychology;
Proceedings of the National Academy of Sciences; Proceedings of
the Society for Experimental Biology and Medicine; Psychological
Review; Quarterly Review of Biology .

Title

Author

Accession No

Call No:

Issue

bate

CONTENTS

CBAFTn

I. The Nature of Scientific Theory .

II. Introduction to an Objective Theory of Be-
havior

III. Stimulus Reception and Organism Survival .

IV. The BiologicaTProblem of Action and Its Co-

ordination . ' \ .

V. Characteristics of Innate Behavior Under Con-
ditions of Need . . ( . . . * .

VI. The Acquisition of Receptor-Effecior Connec-
tions — Primary Reinforcement . . . .

VII. The Acquisition of Receptor-Effector Connec-
tions :: =Secondary Reinforcement . . . .

VIII. The Symbolic Construct bHr as a Function of
the Number~of Reinforcements

IX. Habit Strength as a Function of the Nature and
Amount of the Reinforcing Agent . . .

X. Habit Strength and the Time Interval Sepa-
rating Reaction from Reinforcement .

XI. Habit Strength as a Function of the Temporal
Relation of the Conditioned Stimulus to the
Reaction

XII. Stimulus- Generalization

XIII. Some Functional Dynamics of Compound Con-

ditioned Stimuli

XIV. Primary Motivation and Reaction Potential .

XV. Unadaptive Habits and Experimental Extinc-
tion . . .

XVI. Inhibition and Effective Reaction Potential .

XVII. Behavioral Oscillation

PASS

102

124

135

165

183

204

226

258

277

304

XXI

xxii

CONTENTS

CHAPTER

^VIII. The Reaction Threshold and Response Evoca*

tion

XIX. The Patterning of Stimulus Compounds •„
XX. General Summary and Conclusions . „

Glossary of Symbols

Index of Names

Index of Subjects

PACE

322

349

381

403

411

415

Principles of Behavior

• •

CHAPTER I

The Nature of Scientific Theory

This book is the beginning of an attempt to sketch a systematic
objective theory of the behavior of higher organisms. It is accord-
ingly important at the outset to secure a clear notion of the essential
nature of systematic theory in science, the relation of theory to
other scientific activities, and its general scientific status and
importance.

THE TWO ASPECTS OF SCIENCE: EMPIRICAL AND EXPLANATORY

Men are ever engaged in the dual activity of making observa-
tions and then seeking explanations of the resulting revelations.
All normal men in all times have observed the rising and setting
of the sun and the several phases of the moon. The more thought-
ful among them have then proceeded to ask the question, ‘‘Why?
Why does the moon wax and wane? Why does the sun rise and
set, and where does it go when it sets?" Here we have the two
essential elements of modern science: the making of observations
constitutes the empirical or factual component, and the systematic
attempt to explain these facts constitutes the theoretical com-
ponent. As science has developed, specialization, or division of
labor, has occurred; some men have devoted their time mainly to
the making of observations, while a smaller number have occupied
themselves largely with the problems of explanation.

During the infancy of science, observations are for the most
part casual and qualitative — the sun rises, beats down strongly at
midday, and sets; the moon grows from the crescent to full and
then diminishes. Later observations, usually motivated by prac-
tical considerations of one kind or another, tend to become quan-
titative and precise— the number of days in the moon’s monthly
cycle are counted accurately, and the duration of the sun’s yearly
course is determined with precision. As the need for more exact
observations increases, special tools and instruments, such as gradu-
ated measuring sticks, protractors, clocks, telescopes, and micro-
scopes, are devised to facilitate the labor. Kindred tools relating
to a given field of science are frequently assembled under a single

PRINCIPLES OF BEHAVIOR

roof for convenience of use; such an assemblage becomes a labora-
tory.

As scientific investigations become more and more searching
it is discovered that the spontaneous happenings of nature are not
adequate to permit the necessary observations. This leads to the
setting up of special conditions which will bring about the desired
events under circumstances favorable for such observations; thus
experiments originate. But even in deliberate experiment it is
often extraordinarily difficult to determine with which among a
complex of antecedent conditions a given consequence is primarily
associated; in this way arise a complex maze of control experiments
and other technical procedures, the general principles of which are
common to all sciences but the details of which are peculiar to
each. Thus in brief review we see the characteristic technical
development of the empirical or factual aspect of science.

Complex and difficult as are some of the problems of empirical
science, those of scientific theory are perhaps even more difficult
of solution and are subject to a greater hazard of error. It is not
a matter of chance that the waxing and waning of the moon was
observed for countless millennia before the comparatively recent
times when it was at last successfully explained on the basis of the
Copernican hypothesis. Closely paralleling the development of the
technical aids employed by empirical science, there have also grown
up in the field of scientific theory a complex array of tools and
special procedures, mostly mathematical and logical in nature, de-
signed to aid in coping with these peculiar difficulties. Because of
the elementary nature of the present treatise, very little explicit
discussion of the use of such tools will be given.

THE DEDUCTIVE NATURE OF SCIENTIFIC THEORY AND

EXPLANATION

The term theory in the behavioral or "social” sciences has a
variety of current meanings. As understood in the present work,
a theory is a systematic deductive derivation of the secondary
principles of observable phenomena from a relatively small number
of primary principles or postulates, much as the secondary prin-
ciples or theorems of geometry are all ultimately derived as a logical
hierarchy from a few original definitions and primary principles
called axioms. In science an observed event is said to be explained
when the proposition expressing it has been logically derived from

THE NATURE OF SCIENTIFIC THEORY 3

a set of definitions and postulates coupled with certain observed
conditions antecedent to the event. This, in brief, is the nature
of scientific theory and explanation as generally understood and
accepted in the physical sciences after centuries of successful devel-
opment (f, pp. 495-496).

The preceding summary statement of the nature of scientific
theory and explanation needs considerable elaboration and exem-
plification. Unfortunately the finding of generally intelligible
examples presents serious difficulties; because of the extreme youth
of systematic behavior theory (7, p. 501 flf. ; 2 , p. 15 fl.) as here
understood, it is impossible safely to assume that the reader pos-
sesses any considerable familiarity with it. For this reason it will
be necessary to choose all the examples from such physical sciences
as are now commonly taught in the schools.

We can best begin the detailed consideration of the nature of
scientific explanation by distinguishing it from something often
confused -with it. Suppose a naive person with a moderate-sized
telescope has observed Venus, Mars, Jupiter, and Saturn, together
with numerous moons (including our own), and found them all
to be round in contour and presumably spherical in form. He
might proceed to formulate his observations in a statement such as,
“All heavenly bodies are spherical,” even though this statement
goes far beyond the observations, since he has examined only a
small sample of these bodies. Suppose, next, he secures a better
telescope; he is now able to observe Uranus and Neptune, and finds
both round in contour also. He may, in a manner of speaking, be
said to explain the sphericity of Neptune by subsuming it under
the category of heavenly bodies and then applying his previous
empirical generalization. Indeed, he could have predicted the
spherical nature of Neptune by this procedure before it was observed
at all:

All heavenly bodies are spherical.

Neptune is a heavenly body,

Therefore Neptune is spherical.

Much of what is loosely called explanation in the field of be-
havior is of this nature. The fighting propensities of a chicken
are explained by the fact that he is a game cock and game cocks
are empirically known to be pugnacious. The gregariousness of a
group of animals is explained by the fact that the animals in ques-
tion are dogs, and dogs are empirically known to be gregarious.

PRINCIPLES OF BEHAVIOR

As we have seen, it is possible to make concrete predictions of a
sort on the basis of such generalizations, and so they have signifi-
cance. Nevertheless this kind of procedure — the subsumption of a
particular set of conditions under a category involved in a pre-
viously made empirical generalization — is not exactly what is
regarded here as a scientific theoretical explanation.

For one thing, a theoretical explanation as here understood
grows out of a problem, e.g., “What must be the shape of the
heavenly bodies?” Secondly, it sets out from certain propositions
or statements. These propositions are of two rather different kinds.
Propositions of the first type required by an explanation are those
stating the relevant initial or antecedent conditions. For example,
an explanation of the shape of heavenly bodies might require the
preliminary assumption of the existence of (1) a large mass of
(2) more or less plastic, (3) more or less homogeneous matter,
(4) initially of any shape at all, (5) the whole located in otherwise
empty space. But a statement of the antecedent conditions is noc
enough; there must also be available a set of statements of general
'principles or rules of action relevant to the situation. Moreover,
the particular principles to be utilized in a given explanation must
be chosen from the set of principles generally employed by the
theorist in explanations of this class of phenomena, the choice to
be made strictly on the basis of the nature of the question or
problem under consideration taken in conjunction with the ob-
served or assumed conditions. For example, in the case of the
shape of the heavenly bodies the chief principle employed is the
Newtonian law of gravitation, namely, that every particle of matter
attracts every other particle to a degree proportional to the product
of their masses and inversely proportional to the square of the dis-
tances separating them. These principles are apt themselves to
be verbal formulations of empirical generalizations, but may be
merely happy conjectures or guesses found by a certain amount of
antecedent trial-and-error to agree with observed fact. At all

events they originate in one way or another in empirical observa-
tion.

The concluding phase of a scientific explanation is the deriva-
tion of the answer to the motivating question from the conditions
and the principles, taken jointly, by a process of inference or rea-
soning. For example, it follows from the principle of gravitation
that empty spaces which might at any time have existed within
the mass of a heavenly body would at once be closed. Moreover,

THE NATURE OF SCIENTIFIC THEORY $

if at any point on the surface there were an elevation and adjacent
to it a depression or valley, the sum of the gravitational pressures
of the particles of matter in the elevation acting on the plastic
material beneath would exert substantially the same pressure later-
ally as toward the center of gravity. But since there would be no
equal lateral pressure originating in the valley to oppose the pres-
sure originating in the elevation, the matter contained in the
elevation would flow into the valley, thus eliminating both. This
means that in the course of time all the matter in the mass under
consideration would be arranged about its center of gravity with
no elevations or depressions; i.e., the radius of the body at all
points would be the same. In other words, if the assumed mass were
not already spherical it would in the course of time automatically
become so U, p. 424). It follows that all heavenly bodies, includ-
ing Neptune, must be spherical in form.

The significance of the existence of these two methods of
arriving at a verbal formulation of the shape of the planet Neptune
may now be stated. The critical characteristic of scientific theo-
retical explanation is that it reaches independently through a
process of reasoning the same outcome with respect to (secondary )
principles as is attained through the process of empirical general-
ization. Thus scientific theory may arrive at the general proposi-
tion, “All heavenly bodies of sufficient size, density, plasticity, and
homogeneity are spherical,” as a theorem, simply by means of a
process of inference or deduction without any moons or planets
having been observed at all. The fact that, in certain fields at
least, practically the same statements or propositions can be at-
tained quite independently by empirical methods as by theoretical
procedures is of enormous importance for the development of
science. For one thing, it makes possible the checking of results
obtained by one method against those obtained by the other. It is
a general assumption in scientific methodology that if everything
entering into both procedures is correct, the statements yielded by
them will never be in genuine conflict.

SCIENTIFIC EXPLANATIONS TEND TO COME IN CLUSTERS
CONSTITUTING A LOGICAL HIERARCHY

This brings us to the important question of what happens in
a theoretical situation when one or more of the supposed ante-
cedent conditions are changed, even a little. For example, when

PRINCIPLES OF BEHAVIOR

considering the theoretical shape of heavenly bodies, instead of the
mass being completely fluid it might be assumed to be only slightly
plastic. It is evident at once, depending on the degree of plasticity,
the size of the mass, etc., that there may be considerable deviation
from perfect sphericity, such as the irregularities observable on the
surface of our own planet. Or suppose that we introduce the addi-
tional condition that the planet revolves on its axis. This neces-
sarily implies the entrance into the situation of the principle of
centrifugal force, the familiar fact that any heavy object whirled
around in a circle will pull outward. From this, in conjunction with
other principles, it may be reasoned (and Newton did so reason)
that the otherwise spherical body would bulge at the equator;
moreover, this bulging at the equator together with the principle
of gravity would, in turn, cause a flattening at the poles (4, p. 424).
Thus we see how it is that as antecedent conditions are varied the
theoretical outcome (theorem) following from these conditions will
also vary. By progressively varying the antecedent conditions in
this way an indefinitely large number of theorems may be derived,
but all from the very same group of basic principles. The prin-
ciples are employed over and over in different combinations, one
combination for each theorem. Any given principle may accord-
ingly be employed many times, each time in a different context.
In this way it comes about that scientific theoretical systems po-
tentially have a very large number of theorems (secondary prin-
ciples) but relatively few general (primary) principles.

We note, next, that in scientific systems there are not only many
theorems derived by a process of reasoning from the same assem-
blage of general principles, but these theorems take the form of a
logical hierarchy: first-order theorems are derived directly from the
original general principles; second-order theorems are derived with
the aid of the first-order theorems; and so on in ascending hierarchi-
cal orders. Thus in deducing the flattening of the planets at the
poles, Newton employed the logically antecedent principle of cen-
trifugal force which, while an easily observable phenomenon, can
itself be deduced, and so was deduced by Newton, from the condi-
tions of circular motion. The principle of centrifugal force accord-
ingly is an example of a lower-order theorem in Newton’s theo-
retical system (4, p. 40 ff.). On the other hand, Newton derived
from the bulging of the earth at its equator what is known as the
“precession of the equinoxes” (4, p. 580) , the fact that the length
of the year as determined by the time elapsing from one occasion

THE NATURE uF SCIENTIFIC THEORY 7

when the shadow cast by the winter sun at noon is longest to the
next such occasion, is shorter by some twenty minutes than the
length of the year as determined by noting the time elapsing from
the conjunction of the rising of the sun with a given constellation
of stars to the next such conjunction. This striking phenomenon,
discovered by Hipparchus in the second century b.c., was first ex-
plained by Newton. The precession of the equinoxes accordingly
is an example of a higher-order theorem in the Newtonian theo-
retical system.

From the foregoing it is evident that in its deductive nature
systematic scientific theory closely resembles mathematics. In this
connection the reader may profitably recall his study of geometry
with (1) its definitions, e.g., point, line, surface, etc., (2) its primary
principles (axioms), e.g., that but one straight line can be drawn
between two points, etc., and following these (3) the ingenious and
meticulous step-by-step development of the proof of one theorem
after the other, the later theorems depending on the earlier ones in
a magnificent and ever-mounting hierarchy of derived propositions.
Proper scientific theoretical systems conform exactly to all three of
these characteristics. 1 For example, Isaac Newton’s Principia (4),
the classical scientific theoretical system of the past, sets out with
(1) seven definitions concerned with such notions as matter, mo-
tion, etc., and (2) a set of postulates consisting of his three famous
laws of motion, from which is derived (3) a hierarchy of seventy-
three formally proved theorems together with large numbers of
appended corollaries. The theorems and corollaries are concerned
with such concrete observable phenomena as centrifugal force, the
shape of the planets, the precession of the equinoxes, the orbits of
the planets, the flowing of the tides, and so on.

SCIENTIFIC THEORY IS NOT ARGUMENTATION

The essential characteristics of scientific theory may be further
clarified b^ contrasting it with argumentation and even with
geometry. It is true that scientific theory and argument have
similar formal or deductive structures; when ideally complete both
should have their terms defined, their primary principles stated*

1 The formal structure of scientific theory differs in certain respects from
that of pure mathematics, but these differences need not be elaborated here;
the point to be emphasized is that mathematics and scientific theory are
alike in that they are both strictly deductive in their natures.

PRINCIPLES OF BEHAVIOR

and their conclusions derived in an explicit and logical manner. In
spite of this superficial similarity, however, the two differ radically
in their essential natures, and it would be difficult to make a more
serious mistake than to confuse them. Because of the widespread
tendency to just this confusion, the distinction must be stressed.
An important clue to the understanding of the critical differences
involved is found in the objectives of t^e two processes.

The 'primary objective of argumentation is persuasion. It is
socially aggressive; one person is deliberately seeking to influence
or coerce another by means of a process of reasoning. There is
thus in argumentation a proponent and a recipient. On the surface
the proponent’s objective often appears to be nothing more than to
induce the recipient to assent to some more or less abstract proposi-
tion. Underneath, however, the ultimate objective is usually to
lead the recipient to some kind of action, not infrequently such as
to be of advantage to the proponent or some group with which the
proponent is allied. Now, for the effort involved in elaborate argu-
mentation to have any point, the proposition representing the objec-
tive of the proponent’s efforts must be of such a nature that it
cannot be substantiated by direct observation. The recipient can-
not have made such observations; otherwise he would not need to

be convinced.

Moreover, for an argument to have any coerciveness, the
recipient must believe that the definitions and the other basic
assumptions of the argument are sound ; the whole procedure is that
of systematically transferring to the final culminating conclusion
the assent which the recipient initially gives to these antecedent
statements. In this connection it is to be noted that systems of
philosophy, metaphysics, theology, etc., are in the above sense at
bottom elaborate arguments or attempts at persuasion, since their
conclusions are of such a nature that they cannot
lished by direct observation. Consider, for example, Proposition
XIV of Part One of Spinoza’s Ethic (5) :

possibly be estab-

“Besides God no substance can be, . . .”

The primary objective of scientific theory, on the other hand,
is the establishment of scientific principles. Whereas argumenta-
tion is socially aggressive and is directed at some other person,
natural science theory is aggressive towards the problems of nature,
and it uses logic as a tool primarily for mediating to the scientist
himself a more perfect understanding of natural processes. If New-

THE NATURE OF SCIENTIFIC THEORY

ton had been a scientific Robinson Crusoe, forever cut off from
social contacts, he would have needed to go through exactly the
same logical processes as he did, if he were himself to have under-
stood why the heavenly bodies are spherical rather than cubical.
Naturally also, argumentation presupposes that the proponent has
the solution of the question at issue fully in hand; hence his fre-
quent overconfidence, aggressiveness, and dogmatism. In contrast
to this, the theoretical activities of science, no less than its em-
pirical activities, are directed modestly toward the gradual, piece-
meal, successive- approximation establishment of scientific truths.
In a word, scientific theory is a technique of investigation, of seek-
ing from nature the answers to questions motivating the investi-
gator; it is only incidentally and secondarily a technique of per-
suasion. It should never descend to the level of mere verbal
fencing so characteristic of metaphysical controversy and argu-
mentation.

Some forms of argumentation, such as philosophical and meta-
physical speculation, have often been supposed to attain certainty
of their conclusions because of the “self-evident” nature of their
primary or basic principles. This is probably due to the influence
of Euclid, who believed his axioms to be “self-evident truths.” At
the present time mathematicians and logicians have largely aban-
doned intuition or self-evidentiality as a criterion of basic or any
other kind of truth. Similarly, scientific theory recognizes no
axiomatic or self-evident truths ; it has postulates but no axioms in
the Euclidian sense. Not only this; scientific theory differs sharply
from argumentation in that its postulates are not necessarily sup-
posed to be true at all. In fact, scientific theory largely inverts
the procedure found in argument: whereas argument reaches belief
in its theorems because of antecedent belief in its postulates , scien-
tific theory reaches belief in its postulates to a considerable extent
through direct or observational evidence of the soundness of its

theorems (2, p. 7).

THEORETICAL AND EMPIRICAL PROCEDURES CONTRIBUTE
JOINTLY TO THE SAME SCIENTIFIC END

No doubt the statement that scientific theory attains belief in
its postulates through belief in the soundness of its theorems will
come as a distinct surprise to many persons, and for several reasons.
For one thing, the thoughtful individual may wonder why, in spite

IO PRINCIPLES OF BEHAVIOR

of the admitted absence of self-evident principles or axioms, the
basic principles of scientific systems are not firmly established at
the outset by means of observation and experiment. After such
establishment, it might be supposed, the remaining theorems of
the system could all be derived by an easy logical procedure with-
out the laborious empirical checking of each, as is the scientific
practice. Despite the seductive charm of its simplicity this
methodology is, alas, impossible. One reason is, as already pointed
out (p. 3), that the generalizations made from empirical investi-
gations can never be quite certain. Thus, as regards the purely
empirical process every heavenly body so far observed might be
spherical, yet this fact would only increase the probability that the
next one encountered would be spherical; it would not make it
certain. The situation is exactly analogous to that of the con-
tinued drawing of marbles at random from an urn containing
white marbles and suspected of containing black ones also. As
one white marble after another is drawn in an unbroken succession
the probability increases that the next one drawn will be white, but
there can never come a time when there will not be a margin of
uncertainty. On the basis of observation alone to say that all
heavenly bodies are spherical is as unwarranted as it would be to
state positively that all the marbles in an urn must be white be-
cause a limited random sampling has been found uniformly to be
white.

But even for the sampling of empirical theory or experimental
truth to be effective, the sampling of the different situations in-
volved must be truly random. This means that the generalization
in question must be tried out empirically with all kinds of ante-
cedent conditions; w T hich implies that it must be tested in conjunc-
tion with the operation of the greatest variety of other principles,
singly and in their various combinations. In very simple situations
the scientist in search of primary principles needs to do little more
than formulate his observations. For example, it is simple enough
to observe the falling of stones and similar heavy objects, and
even to note that such objects descend more rapidly the longer the
time elapsed since they were released from rest. But the moment
two or more major principles are active in the same situation, the
task of determining the role played by the one under investigation
becomes far more difficult. It is not at all obvious to ordinary
observation that the principle of gravity operative in the behavior
of freely falling bodies is the same as that operative in the behavior

THE NATURE OF SCIENTIFIC THEORY

of the common pendulum; ordinary falling bodies, for example, do
not manifest the phenomena of lateral oscillation. The relation of
gravity to the behavior of the pendulum becomes evident only as
the result of a fairly sophisticated mathematical analysis requiring
the genius of a Galileo for its initial formulation. But this
“mathematical analysis,” be it noted, is full-fledged scientific theory
with bona fide theorems such as: the longer the suspension of the
pendulum, the slower the beat.

In general it may be said that the greater the number of addi-
tional principles operative in conjunction with the one under inves-
tigation, the more complex the theoretical procedures which are
necessary. It is a much more complicated procedure to show theo-
retically that pendulums should beat more slowly at the equator
than at the pole than it is to deduce that pendulums with long
suspensions should beat more slowly than those with short ones.
This is because in the former situation there must be taken explicitly
into consideration the additional principle of the centrifugal force
due to the rotation of the earth about its axis.

At the outset of empirical generalization it is often impossible
to detect and identify the active scientific principle by mere obser-
vation. For example, Newton's principle that all objects attract
each other inversely as the squares of the distances separating them
was a daring conjecture and one extending much beyond anything
directly observable in the behavior of ordinary falling bodies. It
is also characteristic that the empirical verification of this epoch-
making principle was first secured through the study of careful
astronomical measurements rather than through the observation
of small falling objects. But the action of gravity in determining
the orbits of the planets is even less obvious to ordinary unaided
observation than is its role in the determination of the behavior
of the pendulum. Indeed, this can be detected only by means of
the mathematics of the ellipse, i.e., through a decidedly sophisticated
theoretical procedure, one which had to await the genius of Newton
for its discovery.

Earlier in the chapter it was pointed out that the theoretical
outcome, or theorem, derived from a statement of supposed ante-
cedent conditions is assumed in science always to agree with the
empirical outcome, provided both procedures have been correctly
performed. We must now note the further assumption that if there
is disagreement between the two outcomes there must be something
wrong with at least one of the principles or rules involved in the

PRINCIPLES OF BEHAVIOR

derivation of the theorem; empirical observations are regarded as
primary, and wherever a generalization really conflicts with obser-
vation the generalization must always give way. When the break-
down of a generalization occurs in this way, an event of frequent
occurrence in new fields, the postulates involved are revised if
possible so as to conform to the known facts. Following this, de-
ductions as to the outcome of situations involving still other com-
binations of principles are made; these in their turn are checked
against observations; and so on as long as disagreements continue
to occur. Thus the determination of scientific principles is in con-
siderable part a matter of symbolic trial-and-error. At each trial
of this process, where the antecedent conditions are such as to in-
volve jointly several other presumptive scientific principles, sym-
bolic or theoretical procedure is necessary in order that the investi-
gator may know the kind of outcome to be expected if the supposed
principle specially under investigation is really acting as assumed.
The empirical procedure is necessary in order to determine whether
the antecedent conditions were really followed by the deductively
expected outcome. Thus both theoretical and empirical procedures
are indispensable to the attainment of the major scientific goal —
that of the determination of scientific principles.

HOW THE EMPIRICAL VERIFICATION OF THEOREMS INDIRECTLY

SUBSTANTIATES POSTULATES

But how can the empirical verification of the implications of
theorems derived from a set of postulates establish the truth of the
postulates? In seeking an answer to this question we must note
at the outset that absolute truth is not thus established. The con-
clusion reached in science is not that the postulates employed in
the derivation of the empirically verified theorem are thereby shown
to be true beyond doubt, but rather that the empirical verification
of the theorem has increased the 'probability that the next theorem
derived from these postulates in conjunction with a different set
of antecedent conditions will also agree with relevant empirical
determinations. And this conclusion is arrived at on the basis of
chance or probability, i.e., on the basis of a theory of sampling.

The nature of this sampling theory may be best explained by
means of a decidedly artificial example. Suppose that by some
miracle a scientist should come into possession of a set of postulates
none of which had ever been employed, but which were believed

THE NATURE OF SCIENTIFIC THEORY 1 3

to satisfy the logical criterion of yielding large numbers of em-
pirically testable theorems; that a very large number of such
theorems should be deducible by special automatic logical calcula-
tion machines; that the theorems, each sealed in a neat capsule,
should all be turned over to the scientist at once; and, finally, that
these theorems should then be placed in a box, thoroughly mixed,
drawn out one at a time, and compared with empirical fact. As-
suming that no failures of agreement occurred in a long succession
of such comparisons, it would be proper to say that each succeeding
agreement would increase the probability that the next drawing
from the box would also result in an agreement, exactly as each
successive uninterrupted drawing of white marbles from an urn
would increase progressively the probability that the next drawing
would also yield a white marble. But just as the probability of
drawing a white marble will always lack something of certainty
even with the best conceivable score, so the validation of scientific
principles by this procedure must always lack something of being
complete. Theoretical “truth” thus appears in the last analysis to
be a matter of greater or less probability. It is consoling to know
that this probability frequently becomes very high indeed (3, p. 6) .

THE “ truth ” STATUS OF LOGICAL PRINCIPLES OR RULES

Despite much belief to the contrary, it seems likely that logical
(mathematical) principles are essentially the same in their mode
of validation as scientific principles; they appear to be merely
invented rules of symbolic manipulation which have been found by
trial in a great variety of situations to mediate the deduction of
existential sequels verified by observation. Thus logic in science
is conceived to be primarily a tool or instrument useful for the
derivation of dependable expectations regarding the outcome of
dynamic situations. Except for occasional chance successes, it
requires sound rules of deduction, as well as sound dynamic postu-
lates, to produce sound theorems. By the same token, each obser-
vationally confirmed theorem increases the justified confidence in
the logical rules which mediated the deduction, as well as in the
“empirical” postulates themselves. The rules of logic are more
dependable, and consequently less subject to question, presumably
because they have survived a much longer and more exacting period
of trial than is the case with most scientific postulates. Probably
it is because of the widespread and relatively unquestioned ac-

! 4

PRINCIPLES OF BEHAVIOR

ceptance of the ordinary logical assumptions, and because they
come to each individual investigator ready-made and usually with-
out any appended history, that logical principles are so frequently
regarded with a kind of religious awe as a subtle distillation of the
human spirit; that they are regarded as never having been, and as
never to be, subjected to the tests of validity usually applicable to
ordinary scientific principles; in short, that they are strictly “self-
evident” truths (5, p. 7). As a kind of empirical confirmation of
the above view as to the nature of logical principles, it may be noted
that both mathematicians and logicians are at the present time
busily inventing, modifying, and generally perfecting the principles
or rules of their disciplines ( 6 ).

SUMMARY

Modern science has two inseparable components — the empirical
and the theoretical. The empirical component is concerned pri-
marily with observation; the theoretical component is concerned
with the interpretation and explanation of observation. A natural
event is explained when it can be derived as a theorem by a process
of reasoning from (1) a knowledge of the relevant natural condi-
tions antedating it, and (2) one or more relevant principles called
postulates. Clusters or families of theorems are generated, and
theorems are often employed in the derivation of other theorems;
thus is developed a logical hierarchy resembling that found in
ordinary geometry. A hierarchy of interrelated families of theo-
rems, all derived from the same set of consistent postulates, con-
stitutes a scientific system.

Scientific theory resembles argumentation in being logical in
nature but differs radically in that the objective of argument is to
convince. In scientific theory logic is employed in conjunction
with observation as a means of inquiry. Indeed, theoretical pro-
cedures are indispensable in the establishment of natural laws. The
range of validity of a given supposed law can be determined only
by trying it out empirically under a wide range of conditions where
it will operate in simultaneous conjunction with the greatest variety
and combination of other natural laws. But the only way the
scientist can tell from the outcome of such an empirical procedure
whether a given hypothetical law has acted in the postulated
manner is first to deduce by a logical process what the outcome

THE NATURE OF SCIENTIFIC THEORY 15

of the investigation should be if the hypothesis really holds. This
deductive process is the essence of scientific theory.

The typical procedure in science is to adopt a postulate tenta-
tively, deduce one or more of its logical implications concerning
observable phenomena, and then check the validity of the deduc-
tions by observation. If the deduction is in genuine disagreement
with observation, the postulate must be either abandoned or so
modified that it implies no such conflicting statement. If, however,
the deductions and the observations agree, the postulate gains in
dependability. By successive agreements under a very wide variety
of conditions it may attain a high degree of justified credibility, but
never absolute certainty.

REFERENCES

1. Hull, C. L. The conflicting psychologies of learning— a way out. Psychol.

Rev., 1935, 491-516.

2. Hull, C. L. Mind, mechanism, and adaptive behavior. Psychol. Rev.,

1937, U, 1-32. _ _

3. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. Mathematico-deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

4. Newton, I. Principia (trans. by F. Cajori). Berkeley: Univ. Calif. Press,

1934.

5. Spinoza, B. de. Ethic (trans. by W. H. White and A. H. Stirling). New

York: Macmillan Co., 1894.

6. Whitehead, A. N., and Russell, M. A. Principia mathematica (Vol. I).

London: Cambridge Univ. Press, 1935.

CHAPTER II

Introduction to an Objective Theory of Behavior

Having examined the general nature of scientific theory, we
must now proceed to the elaboration of an objective theory as
applied specifically to the behavior of organisms { 10 ). Preliminary
to this great and complex task it will be well to consider a number
of the more general characteristics of organismic behavior, as well
as certain difficulties which will be encountered and hazards which
ought to be avoided.

THE BASIC FACT OF ENVIRONMENTAI/-ORGANISMIC INTERACTION'

At the outset of the independent life of an organism there
begins a dynamic relationship between the organism and its environ-
ment. For the most part, both environment and organism are
active; the environment acts on the organism, and the organism
acts on the environment (5, p. 2). Naturally the terminal phase of
any given environmental-organismic interaction depends upon the
activity of each; rarely or never can the activity of either be pre-
dicted from knowing the behavior characteristics of one alone.
The possibility of predicting the outcome of such interaction de-
pends upon the fact that both environment and organism are part
of nature, and as such the activity of each takes place according
to known rules, i.e., natural laws.

The environment of an organism may conveniently be divided
into two portions — the internal and the external. The external
environment may usefully be subdivided into the inanimate en-
vironment and the animate or organismic environment.

The laws of the internal environment are, for the most part,
those of the physiology of the particular organism. The laws of
the inanimate environment are those of the physical world and
constitute the critical portions of the physical sciences; they are
relatively simple and reasonably well known.

The laws of the organismic environment are those of the be-
havior of other organisms, especially organisms of the same species
as the one under consideration; they make up the primary prin-
ciples of the behavior, or “social,” sciences and are comparatively

NATURE OF OBJECTIVE BEHAVIOR THEORY 17

complex. Perhaps because of this complexity they are not as yet
very well understood. Since in a true or symmetrical social situa-
tion only organisms of the same species are involved, the basic laws
of the activities of the environment must be the same as those of
the organism under consideration. It thus comes about that the
objective of the present work is the elaboration of the basic molar 1
behavioral laws underlying the “social" sciences.

ORGANISMIC NEED, ACTIVITY, AND SURVIVAL

Since the publication by Charles Darwin of the Origin of Species
(2) it has been necessary to think of organisms against a back-
ground of organic evolution and to consider both organismic struc-
ture and function in terms of survival. Survival, of course, applies
equally to the individual organism and to the species. Physio-
logical studies have shown that survival requires special circum-
stances in considerable variety; these include optimal conditions of
air, water, food, temperature, intactness of bodily tissue, and so
forth; for species survival among the higher vertebrates there is
required at least the occasional presence and specialized reciprocal
behavior of a mate.

On the other hand, when any of the commodities or conditions
necessary for individual or species survival are lacking, or when
they deviate materially from the optimum, a state of primary need
is said to exist. In a large proportion of such situations the need
will be reduced or eliminated only through the action on the en-
vironment of a particular sequence of movements made by the
organism. For example, the environment will, as a rule, yield a
commodity (such as food) which will mediate the abolition of a
state of need (such as hunger) only when the movement sequence
corresponds rather exactly to the momentary state of the environ-
ment; i.e., when the movement sequence is closely synchronized
with the several phases of the environmental reactions. If it is to
be successful, the behavior of a hungry cat in pursuit of a mouse
must vary from instant to instant, depending upon the movements
of the mouse. Similarly if the mouse is to escape the cat, its move-

*By this expression is meant the uniformities discoverable among the
grossly observable phenomena of behavior as contrasted with the laws of the
behavior of the ultimate “molecules” upon which this behavior depends,
such as the constituent cells of nerve, muscle, gland, and so forth. The term
molar thus means coarse or macroscopic as contrasted with molecular, or
microscopic.

1 8 PRINCIPLES OF BEHAVIOR

ments must vary from instant to instant, depending upon the move-
ments of the cat.

Moreover, in a given external environment situation the be-
havior must often differ radically from one occasion to another,
depending on the need which chances to be dominant at the time;
e.g., whether it be of food, water, or a mate. In a similar manner
the behavior must frequently differ widely from one environmental
situation to another, even when the need is exactly the same in each
environment; a hungry man lost in a forest must execute a very
different sequence of movements to relieve his need from what
would be necessary if he were in his home.

It follows from the above considerations that an organism will
hardly survive unless the state of organismic need and the state of
the environment in its relation to the organism are somehow jointly
and simultaneously brought to bear upon the movement-producing
mechanism of the organism.

THE ORGANIC BASIS OF ADAPTIVE BEHAVIOR

All normal higher organisms possess a great assortment of
muscles, usually with bony accessories. These motor organs are
ordinarily adequate to mediate the reduction of most needs, pro-
vided their contractions occur in the right amount, combination, and
sequence. The momentary status of most portions of the environ-
ment with respect to the organism is mediated to the organism by
an immense number of specialized receptors which respond to a
considerable variety of energies such as light waves (vision), sound
waves (hearing), gases (smell), chemical solutions (taste), me-
chanical impacts (touch), and so on. The state of the organism
itself (the internal environment) is mediated by another highly
specialized series of receptors. It is probable that the various
conditions of need also fall into this latter category; i.e., in one
way or another needs activate more or less characteristic receptor
organs much as do external environmental forces.

Neural impulses set in motion by the action of these receptors
pass along separate nerve fibers to the central ganglia of the
nervous system, notably the brain. The brain, which acts as a
kind of automatic switchboard, together with the remainder of the
central nervous system, routes and distributes the impulses to indi-
vidual muscles and glands in rather precisely graded amounts and
sequences. When the neural impulse reaches an effector organ

NATURE OF OBJECTIVE BEHAVIOR THEORY

(muscle or gland) the organ ordinarily becomes active, the amount
of activity usually varying with the magnitude of the impulse.
The movements thus brought about usually result in the elimination
of the need, though often only after numerous unsuccessful trials.
But organismic activity is by no means always successful; not in-
frequently death occurs before an adequate action sequence has
been evoked.

It is the primary task of a molar science of behavior to isolate
the basic laws or rules according to which various combinations of
stimulation, arising from the state of need on the one hand and
the state of the environment on the other, bring about the kind of
behavior characteristic of different organisms. A closely related
task is to understand why the behavior so mediated is so generally
adaptive, i.e., successful in the sense of reducing needs and facili-
tating survival, and why it is unsuccessful on those occasions when
survival is not facilitated.

THE NEUROLOGICAL. VERSUS THE MOLAR APPROACH

From the foregoing considerations it might appear that the
science of behavior must at bottom be a study of physiology. In-
deed, it was once almost universally believed that the science of
behavior must wait for its useful elaboration upon the development
of the subsidiary science of neurophysiology. Partly as a result of
this belief, an immense amount of research has been directed to
the understanding of the detailed or molecular dynamic laws of
this remarkable automatic structure. A great deal has been re-
vealed by these researches and the rate of development is constantly
being accelerated by the discovery of new and more effective
methods of investigation. Nearly all serious students of behavior
like to believe that some day the major neurological laws will be
known in a form adequate to constitute the foundation principles
of a science of behavior.

In spite of these heartening successes, the gap between the
minute anatomical and physiological account of the nervous system
as at present known and what would be required for the construc-
tion of a reasonably adequate theory of molar behavior is im-
passable. The problem confronting the behavior theorist is sub-
stantially like that which would have been faced by Galileo and
Newton had they seriously considered delaying their preliminary
formulation of the molar mechanics of the physical world until the

PRINCIPLES OF BEHAVIOR

micro-mechanics of the atomic and subatomic world had been
satisfactorily elaborated.

Students of the social sciences are presented with the dilemma
of waiting until the physico-chemical problems of neurophysiology
have been adequately solved before beginning the elaboration of
behavior theory, or of proceeding in a provisional manner with
certain reasonably stable principles of the coarse, macroscopic or
molar action of the nervous system whereby movements are evoked
by stimuli, particularly as related to the history of the individual
organism.

There can hardly be any doubt that a theory of molar behavior
founded upon an adequate knowledge of both molecular and molar
principles would in general be more satisfactory than one founded
upon molar considerations alone. But here again the history of
physical science is suggestive. Owing to the fact that Galileo and
Newton carried out their molar investigations, the world has had
the use of a theory which was in very close approximation to obser-
vations at the molar level for nearly three hundred years before
the development of the molecular science of modern relativity and
quantum theory. Moreover, it is to be remembered that science
proceeds by a series of successive approximations; it may very well
be that had Newton’s system not been worked out when it was
there would have been no Einstein and no Planck, no relativity and
no quantum theory at all. It is conceivable that the elaboration
of a systematic science of behavior at a molar level may aid in the
development of an adequate neurophysiology and thus lead in the
end to a truly molecular theory of behavior firmly based on
physiology.

It happens that a goodly number of quasi-neurological principles
have now been determined by careful experiments designed to trace
out the relationship of the molar behavior of organisms, usually
as integrated wholes, to well-controlled stimulus situations. Many
of the more promising of these principles were roughly isolated in
the first instance by the Russian physiologist, Pavlov, and his
pupils, by means of conditioned-reflex experiments on dogs. More
recently extensive experiments in many laboratories in this country
with all kinds of reactions on a wide variety of organisms, includ-
ing man, have greatly extended and rectified these principles and
shown how they operate jointly in the production of the more
complex forms of behavior. Because of the pressing nature of
behavior problems, both practical and theoretical aspects of be-

Kauhtnlr OnWerMW

NATURE OF OBJECTIVE BEHAVIOR THEORY

havior science are, upon the whole, being developed according to the
second of the two alternatives outlined above. For these reasons
the molar approach is employed in the present work.

In this connection it is to be noted carefully that the alternatives
of microscopic versus macroscopic , and molecular versus molar, are
relative rather than absolute. In short, there are degrees of the
molar, depending on the coarseness of the ultimate causal segments
or units dealt with. Other things equal, it would seem wisest to
keep the causal segments small, to approach the molecular, the
fine and exact substructural details, just as closely as the knowledge
of that substructure renders possible. There is much reason to
believe that the seeming disagreements among current students of
behavior may be largely due to the difference in the degree of the
molar at which the several investigators are working. Such dif-
ferences, however, do not represent, fundamental disagreements.
In the end the work of all who differ only in this sense may find
a place in a single systematic structure, the postulates or primary
assumptions of those working at a more molar level ultimately
appearing as theorems of those working at a more molecular level.

THE ROLE OF INTERVENING VARIABLES IN BEHAVIOR THEORY

Wherever an attempt is made to penetrate the invisible world
of the molecular, scientists frequently and usefully employ logical
constructs, intervening variables, or symbols to facilitate their
thinking. These symbols or X’s represent entities or processes
which, if existent, would account for certain events in the observ-
able molar world. Examples of such postulated entities in the field
of the physical sciences are electrons, protons, positrons, etc. A
closely parallel concept in the field of behavior familiar to everyone
is that of habit as distinguished from habitual action. The habit
presumably exists as an invisible condition of the nervous system
quite as much when it is not mediating action as when habitual
action is occurring; the habits upon which swimming is based are
just as truly existent when a person is on the dance floor as when
he is in the water.

In some cases there may be employed in scientific theory a
whole series of hypothetical unobserved entities; such a series is
presented by the hierarchy of postulated physical entities: mole-
cule, atom, and electron, the molecule supposedly being constituted
of atoms and the atom in its turn being constituted of electrons.

PRINCIPLES OF BEHAVIOR

A rough parallel to this chain of hypothetical entities from the
physical sciences will be encountered in the present system of be-
havior theory. For the above reasons the subject of symbolic con-
structs, intervening variables, or hypothetical entities which are not
directly observable requires comment ( 6 , p. 3 ff.).

Despite the great value of logical constructs or intervening
variables in scientific theory, their use is attended with certain
difficulties and even hazards. At bottom this is because the pres-
ence and amount of such hypothetical factors must always be
determined indirectly. But once (1) the dynamic relationship
existing between the amount of the hypothetical entity ( X ) and
some antecedent determining condition (A) which can be directly
observed, and (2) the dynamic relationship of the hypothetical
entity to some third consequent phenomenon or event ( B ) which
also can be directly observed, become fairly well known, the scien-
tific hazard largely disappears. The situation in question is repre-
sented in Figure 1. When a hypothetical dynamic entity, or even

A / >(*) / >B

Fio. 1. Diagrammatic representation of a relatively simple case of an in-
tervening variable (X) not directly observable but functionally related (/) to
the antecedent event ( A ) and to the consequent event ( B ), both A and B
being directly observable. When an intervening variable is thus securely
anchored to observables on both sides it can be safely employed in scientific
theory.

a chain of such entities each functionally related to the one logically
preceding and following it, is thus securely anchored on both sides
to observable and measurable conditions or events (A and B ), the
main theoretical danger vanishes. This at bottom is because under
the assumed circumstances no ambiguity can exist as to when, and
how much of, B should follow A.

THE OBJECTIVE VERSUS THE SUBJECTIVE APPROACH TO

BEHAVIOR THEORY

If the circumstances sketched above as surrounding and safe-
guarding the use of hypothetical entities are not observed, the
grossest fallacies may be committed. The painfully slow path
whereby man has, as of yesterday, begun to emerge into the truly
scientific era is littered with such blunders, often tragic in their

NATURE OF OBJECTIVE BEHAVIOR THEORY 23

practical consequences. A pestilence or a hurricane descends upon
a village and decimates the population. The usual hypothesis put
forward by primitive man (and many others who think themselves
not at all primitive) to explain the tragic event ( B ) is that some
hypothetical spirit (X) has been angered by the violation (A) of
some tribal taboo on the part of one or more inhabitants of the
village. Unfortunately this mode of thinking is deeply ingrained
in most cultures, not excepting our own, and it even crops up under
various disguises in what purports to be serious scientific work.

Perhaps as good an example of such a fallacious use of the
intervening variable as is offered by recent scientific history is that
of the entelechy put forward by Hans Driesch as the central con-
cept in his theory of vitalism (5). Driesch says, for example:

A supreme mind, conversant with the inorganic facts of nature and
knowing all the intensive manifoldness of all entelechies and psychoids . . .
would be able to predict the individual history of the latter, would be able
to predict the actions of any psychoid with absolute certainty. Human
mind, on the other hand, is not able to predict in this way, as it does not
know entelechy before its manifestation, and as the material conditions
of life, which alone the mind of man can know ... in its completeness, are
not the only conditions responsible for organic phenomena. (5, p. 249.)

Driesch’s entelechy (X) fails as a logical construct or intervening
variable not because it is not directly observable (though of course
it is not), but because the general functional relationship to ante-
cedent condition A and that to consequent condition B are both
left unspecified. This, of course, i? but another way of saying that
the entelechy and all similar constructs are essentially metaphysical
in nature. As such they have no place in science. Science has no
use for unverifiable hypotheses.

A logically minded person, unacquainted with the unscientific
foibles of those who affect the scientific virtues, may naturally
wonder how such a formulation could ever mediate a semblance of
theoretical prediction and thus attain any credence as a genuinely
scientific theory. The answer seems to lie in the inveterate animis-
tic or anthropomorphic tendencies of human nature. The entelechy
is in substance a spirit or daemon, a kind of vicarious ghoet. The
person employing the entelechy in effect says to himself, “If I were
the entelechy in such and such a biological emergency, what would
I do?” Knowing the situation and what is required to meet the
emergency, he simply states what he knows to be required as a
solution, and he at once has in this statement what purports to be

24 PRINCIPLES OF BEHAVIOR

a scientific deduction! He has inadvertently substituted himself
in place of the construct and naively substituted his knowledge of
the situation for the objective rules stating the functional rela-
tionships which ought to subsist between A and X on the one hand,
and between X and B on the other.

This surreptitious substitution and acceptance of one’s knowl-
edge of what needs to be done in a biological emergency for a
theoretical deduction is the essence of what we shall call anthro-
pomorphism , or the subjective , in behavior theory. After many
centuries the physical sciences have largely banished the subjective
from their fields, but for various reasons this is far less easy of
accomplishment and is far less well advanced in the field of be-
havior. The only known cure for this unfortunate tendency to which
all men are more or less subject is a grim and inflexible insistence
that all deductions take place according to the explicitly formulated
rules stating the functional relationships of A to X and of X to B.
This latter is the essence of the scientifically objective. A genu-
inely scientific theory no more needs the anthropomorphic intuitions
of the theorist to eke out the deduction of its implications than an
automatic calculating machine needs the intuitions of the operator
in the determination of a quotient, once the keys representing the
dividend and the divisor have been depressed.

Objective scientific theory is necessary because only under ob-
jective conditions can a principle be tested for soundness by means
of observation. The basic difficulty with anthropomorphic sub-
jectivism is that what appear to be deductions derived from such
formulations do not originate in rules stating postulated functional
relationships, but rather in the intuitions of the confused thinker.
Observational check of such pseudo-deductions may verify or refute
these intuitions, but has no bearing on the soundness of any scien-
tific principles whatever; such verifications or refutations might
properly increase the reputation for accurate prophecy of the one
making such intuitive judgments, but a prophet is not a principle,
much less a scientific theory.

OBJECTIVISM VERSUS TELEOLOGY

Even a superficial study of higher organisms shows that their
behavior occurs in cycles. The rise of either a primary or a sec-
ondary need normally marks the beginning of a behavior cycle,
and the abolition or substantial reduction of that need marks its

2 5

NATURE OF OBJECTIVE BEHAVIOR THEORY

end. Some phase of the joint state of affairs resulting from the
environmental-organismic interaction at the end of a behavior cycle
is customarily spoken of as a goal. Our usual thoughtless custom
is to speak of cycles of behavior by merely naming their outcome,
effect, or end result, and practically to ignore the various move-
ments which brought this terminal state about. Guthrie has ex-
pressed this tendency more aptly than anyone else ( 4 , p. 1)- We
say quite naturally that a man catches a fish, a woman bakes a
cake, an artist paints a picture, a general wins a battle. The end
result of each angling exploit, for example, may be in some sense
the same but the actual movements involved are perhaps never
exactly the same on any two occasions; indeed, neither the angler
nor perhaps anyone else knows or could know in their ultimate
detail exactly what movements were made. It is thus inevitable
that for purposes of communication we designate behavior se-
quences by their goals.

Now for certain rough practical purposes the custom of naming
action sequences by their goals is completely justified by its con-
venience. It may even be that for very gross molar behavior it
can usefully be employed in theory construction, provided the
theorist is alert to the naturally attendant hazards. These appear
the moment the theorist ventures to draw upon his intuition for
statements concerning the behavior (movements) executed by the
organism between the onset of a need and its termination through
organismic action. Pseudo-deductions on the basis of intuition bom
cf intimate knowledge are so easy and so natural that the tendency
to make them is almost irresistible to most persons. The practice
does no harm if the theorist does not mistake this subjective intui-
tional performance for a logical deduction from an objective theory,
and attribute the success of his intuitions to the validity of the
theoretical principles.

An ideally adequate theory even of so-called purposive behavior
ought, therefore, to begin with colorless movement and mere
receptor impulses as such, and from these build up step by step
both adaptive behavior and maladaptive behavior. The present
approach does not deny the molar reality of purposive acts (as
opposed to movement), of intelligence, of insight, of goals, of
intents, of strivings, or of value; on the contrary, we insist upon
the genuineness of these forms of behavior. We hope ultimately to
show the logical right to the use of such concepts by deducing them
as secondary principles from more elementary objective primary

2 6

PRINCIPLES OF BEHAVIOR

principles. Once they have been derived we shall not only under-
stand them better but be able to use them with more detailed
effectiveness, particularly in the deduction of the movements which
mediate (or fail to mediate) goal attainment, than would be the
case if we had accepted teleological sequences at the outset as
gross, unanalyzed (and unanalyzable) wholes.

“emergentism” a doctrine of despair

Perhaps the very natural and economical mode of communica-
tion whereby we speak of the terminal or goal phases of action,
largely regardless of the antecedent movements involved, predis-
poses us to a belief in teleology. In its extreme form teleology is
the name of the belief that the terminal stage of certain environ-
mental-organismic interaction cycles somehow is at the same time
one of the antecedent determining conditions which bring the be-
havior cycle about. This approach, in the case of a purposive
behavior situation not hitherto known to the theorist, involves a
kind of logical circularity: to deduce the outcome of any be-
havioral situation in the sense of the deductive predictions here
under consideration, it is necessary to know all the relevant ante-
cedent conditions, but these cannot be determined until the be-
havioral outcome has been deduced. In effect this means that the
task of deduction cannot begin until after it is completed 1
Naturally this leaves the theorist completely helpless. It is not
surprising that the doctrine of teleology leads to theoretical despair
and to such pseudo-remedies as vitalism and emergentism.

Emergentism, as applied to organismic behavior, is the name
for the view that in the process of evolution there has “emerged”
a form of behavior which is ultimately unanalyzable into logically
more primitive elements — behavior which cannot possibly be de-
duced from any logically prior principles whatever. In particular
it is held that what is called goal or purposive behavior is of such
a nature, that it cannot be derived from any conceivable set of
postulates involving mere stimuli and mere movement (5, pp. 7-8;
7, pp. 26-27).

On the other hand, many feel that this defeatist attitude is not
only unwholesome in that it discourages scientific endeavor, but
that it is quite unjustified by the facts. The present writer shareo
this view. Therefore a serious attempt will ultimately be made to
show that these supposedly impossible derivations are actually pos-

NATURE OF OBJECTIVE BEHAVIOR THEORY

sible ; in some cases they will be shown to be quite easy of accom-
plishment.

A SUGGESTED PROPHYLAXIS AGAINST ANTHROPOMORPHIC

SUBJECTIVISM

As already suggested, one of the greatest obstacles to the
attainment of a genuine theory of behavior is anthropomorphic
subjectivism. At bottom this is because we ourselves are so inti-
mately involved in the problem; we are so close to it that it is
difficult to attain adequate perspective. For the reader who has
not hitherto struggled with the complex but fascinating problems
of behavior theory, it will be hard to realize the difficulty of main-
taining a consistently objective point of view. Even when fully
aware of the nature of anthropomorphic subjectivism and its dan-
gers, the most careful and experienced thinker is likely to find
himself a victim to its seductions. Indeed, despite the most con-
scientious effort to avoid this it is altogether probable that there
may be found in various parts of the present work hidden elements
of the anthropomorphically subjective.

One aid to the attainment of behavioral objectivity is to think
in terms of the behavior of subhuman organisms, such as chim-
panzees, monkeys, dogs, cats, and albino rats. Unfortunately this
form of prophylaxis against subjectivism all too often breaks down
when the theorist begins thinking what he would do if he were a
rat, a cat, or a chimpanzee; when that happens, all his knowledge
of his own behavior, born of years of self-observation, at once
begins to function in place of the objectively stated general rules
or principles which are the proper substance of science.

A device much employed by the author has proved itself to be
a far more effective prophylaxis. This is to regard, from time to
time, the behaving organism as a completely self-maintaining robot,
constructed of materials as unlike ourselves as may be. In doing
this it is not necessary to attempt the solution of the detailed
engineering problems connected with the design of such a creature.
It is a wholesome and revealing exercise, however, to consider the
various general problems in behavior dynamics which must be
solved in the design of a truly self-maintaining robot. We, in
common with other mammals, perform innumerable behavior adap-
tations with such ease that it is apt never to occur to us that any
problem of explanation exists concerning them. In many 6uch

PRINCIPLES OF BEHAVIOR

seemingly simple activities lie dynamical problems of very great
complexity and difficulty.

A second and closely related subjective tendency against which
the robot concept is likely to prove effectively prophylactic is that
to the reification of a behavior function. To reify a function is to
give it a name and presently to consider that the name represents
a thing, and finally to believe that the thing so named somehow
explains the performance of the function. We have already seen
an example of this unfortunate tendency in Driesch’s entelechy.
The temptation to introduce an entelechy, soul, spirit, or daemon
into a robot is slight; it is relatively easy to realize that the intro-
duction of an entelechy would not really solve the problem of design
of a robot because there would still remain the problem of designing
the entelechy itself , which is the core of the original problem all
over again. The robot approach thus aids us in avoiding the very
natural but childish tendency to choose easy though false solutions
to our problems, by removing all excuses for not facing them
squarely and without evasion.

Unfortunately it is possible at present to promise an explanation
of only a portion of the problems encountered in the infinitely
complex subject of organismic behavior. Indeed, it is no great
exaggeration to say that the present state of behavior theory re-
sembles one of those pieces of sculpture which present in the main
a rough, unworked block of stone with only a hand emerging in
low relief here, a foot or thigh barely discernible there, and else-
where a part of a face. The undeveloped state of the behavior
sciences suggested by this analogy is a source of regret to the
behavior theorist but not one of chagrin, because incompleteness
is characteristic even of the most advanced of all theoretical
sciences. From this point of view the difference between the physi-
cal and the behavioral sciences is one not of kind but of degree —
of the relative amount of the figure still embedded in the unhewn
rock. There is reason to believe that the relative backwardness
of the behavior sciences is due not so much to their inherent com-
plexity as to the difficulty of maintaining a consistent and rigorous
objectivism.

SUMMARY

The field of behavior theory centers primarily in the detailed
interaction of organism and environment. The basic principles of
organismic behavior are to be viewed against a background of

NATURE OF OBJECTIVE BEHAVIOR THEORY

organic evolution, the success or failure of the evolutionary process
being gauged in terms of survival. Individual and species survival
depend upon numerous optimal physiological conditions; when one
of these critical conditions deviates much from the optimum, a
state of primary need arises. Need reduction usually comes about
through a particular movement sequence on the part of the or-
ganism. Such sequences depend for their success jointly upon the
nature of the need and the nature and state of the environment.

The condition of organismic need and the status of the environ-
ment evoke from specialized receptors neural impulses which are
brought to bear jointly on the motor organs by the central ganglia
of the nervous system acting as an automatic switchboard. The
primary problem of behavior theory is to discover the laws accord-
ing to which this extraordinarily complex process occurs. Students
of behavior have resorted to the coarse, or “molar,” laws of neural
activity as revealed by conditioned-reflex and related experiments,
rather than to the “molecular” results of neurophysiology, because
the latter are not yet adequate.

Perhaps partly as the result of this molar approach, it is found
necessary to introduce into behavior theory numerous logical con-
structs analogous to molecules and atoms long used in the physical
sciences. All logical constructs present grave theoretical hazards
when they are not securely anchored to directly observable events
both as antecedents and consequences by definite functional rela-
tionships. Under conditions of unstated functional relationships
the naive theorist is tempted to make predictions on the basis of
intuition, which is anthropomorphic subjectivism. The derivation
of theoretical expectations from explicitly stated functional rela-
tionships is the objective method. Experimental agreement with
expectations can properly validate theoretical principles only when
objective procedures are employed.

Some writers believe that there is an impassable theoretical
gulf between mere muscle contraction and the attainment of goals;
that the latter are “emergents.” This doctrine of despair grows
naturally out of the doctrine of teleology. The present treatise
accepts neither teleology nor its pessimistic corollary. Goals, in-
tents, intelligence, insight, and value are regarded not only as
genuine but as of the first importance. Ultimately an attempt will
be made to derive all of these things objectively as secondary
phenomena from more elementary objective conditions, concepts,
and principles.

PRINCIPLES OF BEHAVIOR

NOTES

Operational Definitions and Intervening Variables

In 1938 Bridgman, a physicist whose chief research activities have been con-
cerned with the empirical determination of various physical phenomena under
very great pressures, wrote a book (2) in which he made an acute examination
of the use of various concepts in current physical theory, particularly those
representing intervening variables. The cure which he recommended for such
abuses as he found was the scrupulous recognition of the operations carried out
by the experimentalists as a means to the making of the observations and measure-
ments of the observable events (A and B, Figure 1). This, as we saw above, has
special significance for the science of behavior, which is so prone to the subjective
use of intervening variables. Quite naturally and properly, Bridgman^ work
has greatly impressed many psychologists. Unfortunately his emphasis upon
the operations which are the means whereby the observations and measurements
in question become possible has led many psychologists to mistake the means
for the end. The point here to be emphasized is that while observations must be
considered in the context of the operations which make them possible, the central
factor in the situation is what is observed. The moral of Bridgman’s treatise is
that the intervening variable (X) is never directly observed but is an inference
based on the observation of something else, and that the inference is critically
dependent upon the experimental manipulations (operations) which lead to the
observations. An emphasis on operations which ignores the central importance
of the dependent observations completely misses the virtue of what is coining to
be known as operationism.

The Subjective Versus the Objective in Behavior Theory

The critical characteristic of the subjective as contrasted with the objective
is that the subjective tends to be a private event, whereas the objective is a public
event, i.e., an event presumed to be independently observable by many persons.
Thus the perceptual experience or conscious feeling of a person when stimulated
by light rays of a certain wave length is said to be a private or subjective event,
whereas the light rays themselves, or the overt behavior of another person in
response to the impact of the light rays, is said to be a public or objective event.

A typical case of subjectivism in the field of theory, on the other hand, is one
in which the alleged theorist asserts, and even believes, that he has deduced a
proposition in a logical manner, whereas in fact he has arrived at it by mere
anthropomorphic intuition. The subjectivism of behavior theory is thus de-
pendent upon a kind of privacy, but one quite different from that of perceptual
consciousness or experience. The subjective aspect of experience is dependent
upon the private nature of the process hidden within the body of the subject; sub-
jectivism in the field of behavior theory, on the other hand, is dependent upon
the private nature of the processes within the body of the theorist, whereby he
attempts to explain the behavior of the subject. A theory becomes objective
when the primary assumptions and the logical steps whereby these assumptions
lead to further propositions (theorems) are exhibited to public observation and
so make possible a kind of repetition of the logical process by any other person.

NATURE OF OBJECTIVE BEHAVIOR THEORY 3 1

Propositions originating in private intuitions masquerading as unstated logical
processes are, of course, not theoretical material at all, and have no proper place
in science.

Historical Note Concerning the Concept of Molar Behavior

and of the Intervening Variable

The important concept of molar, as contrasted with molecular, behavior was
introduced into psychology in 1931 by E. C. Tolman. The present writer has
taken over the concept substantially as it appears in Tolman ’s well-known
book (8).

The explicit introduction into psychology of the equally important concept
of the intervening variable is also due to Professor Tolman; its first and best
elaboration was given in his address as President of the American Psychological
Association, delivered at Minneapolis, September 3, 1937 (5).

REFERENCES

1. Bridgman, P. W. The logic of modem physics. New York: Macmillan,

1938.

2. Darwin, C. Origin of species. New York: Modem Library, 1936.

3. Driesch, H. The science and philosophy of the organism, second ed.

London: A. and C. Black, 1929.

4. Guthrie, E. R. Association and the law of effect. Psychol. Rev., 1940,

47, 127-148.

5. Hull, C. L. Conditioning: outline of a systematic theory of learning

(Chapter II in The psychology of learning, Forty-First Yearbook of the

National Society for the Study of Education, Part II). Bloomington,

111.: Public School Publishing Co., 1942.

6. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. Mathematico-deductive theory of rote learning* New

Haven: Yale Univ. Press, 1940.

7. Koffka, K. Principles of Gestalt psychology. New York: Harcourt,

Brace and Co., 1935.

8. Tolman, E. C. Purposive behavior in animals and men. New York:

Century Co., 1932.

9. Tolman, E. C. The determiners of behavior at a choice point. Psychol.

Rev., 1938, 1-41.

10. Watson, J. B. Psychology from the standpoint of a behaviorist, second

ed. Philadelphia: J. B. Lippincott Co., 1924.

CHAPTER III

Stimulus Reception and Organism Survival

The intimate relation of the motility of organisms to survival
has repeatedly been emphasized (p. 17). It has also been pointed
out (p. 18) that if the action of organisms is to facilitate survival,
movement must vary in an intimate manner not only with the state
of the need but with the exact state of both the internal and the
external environment at the instant of action occurrence. This
means that the survival of the organism usually requires a precise
integration of the animal’s motor organs with its environment.
The means whereby the various stimulating energies of the environ-
ment are mediated to the nervous system, the central integrating
mechanism of organisms, must therefore be examined.

SOME TERMINOLOGICAL CLARIFICATIONS

Owing in part to the historical contamination of behavior
science by metaphysical speculation, certain ambiguities and mis-
understandings have arisen concerning the meanings of terms com-
monly employed to indicate phenomena associated with receptor
activity and functioning. We shall accordingly state our own use
of these terms, employing the visual receptor for purposes of illus-
tration.

Light is believed by most physicists to be a wave phenomenon,
the different wave lengths (or frequencies) being reflected from the
surfaces of objects. Consider, for example, the large, red celluloid
die represented in Figure 2, against a background of black velvet.
The die itself we shall call a stimulus object . From the surface of
this stimulus object, wave lengths of approximately 650 milli-
microns, say, are reflected in all directions except where opaque
obstructions exist. From the white spots, waves of all lengths are
reflected. The amount of the light reflected in a given direction
will vary jointly with the angle of the surface involved and the
direction of the light source. Any and all of the rays of light re-
flected from the die are potential stimuli , depending upon whether
or not a responsive receptor chances to be in such a position as to

STIMULUS RECEPTION-ORGANISM SURVIVAL 33

receive them; in the latter case the light ray becomes an actual
stimulus (see Figure 2).

Suppose, now, that the die were to be rotated slowly on its
vertical axis as it stands in Figure 2. It is clear that the retina

POTENTIAL
® j STIMULI

,^\SENSE ORGAN
OF OBSERVER

Fia. 2. Diagrammatic representation of a stimulus object, a sheaf of
potential stimuli, an actual stimulus (Si), and the sense organ of the observer
upon which the actual stimulus impinges (5).

would receive the impact of a gradually changing pattern or com-
bination of light waves, each of the infinite number of angles re-
sulting from the rotation reflecting a different configuration of
stimulated points on the retina. Such a series of compound stimu-
lations is called a stimulus continuum.

SOME TYPICAL RECEPTORS AND THEIR ADAPTIVE FUNCTIONS

The processes of organic evolution have solved the primary prob-
lem of mediating to the organism the differential nature and state
of the environment by developing specially differentiated organs
each of which is normally and primarily stimulated only by a
limited range of environmental energy. The normal individuals
of the higher mammalian species possess receptors which respond
to the stimulation of most forms of energy critically active during
the period of evolution. Generally speaking, a distinct receptor
organ is available for each energy type. In this way the several
vitally important energy manifestations of the environment are
qualitatively differentiated. We proceed now to a brief survey of
the more important of these receptors, with special reference to the
biological function performed by each.

Biologically, one of the most primitive and necessary of the
receptors is that responsive to contact or mechanical pressure.

34 PRINCIPLES OF BEHAVIOR

The end organs are appropriately located in the skin or mem-
branes susceptible to contact. An interesting extension of the con-
tact receptor system beyond the surface of the organism is brought
about by the fact that on the hairy parts the touch receptor is
placed at the base of the hairs in such a way that when the shaft
of hair is moved the receptor is activated. Since an object is likely
to touch a hair before it reaches the skin, we have here a beginning
of distance reception. This adaptive device is developed still
further in certain organisms such as cats and rats in the long and
relatively stiff hairs which extend outward from the face, mostly
in the region of the nose. These extended or indirect touch recep-
tors become very useful distance receptors in darkness, even though
their range be small.

One of the many necessities of the mammalian body is the
maintenance of an optimum temperature. This means that recep-
tors must be provided which will yield differential neural impulses
when the environment is such as to cool the skin below, or to
raise it above, a certain temperature. Instead of giving us a single
thermometer to act on the thermostat principle, nature seems to
have evolved two receptors, one for temperatures above about
33.5 C. degrees, and another for temperatures below about 32.5 C.
degrees, depending somewhat upon circumstances. The receptor
organs for temperature have not yet been determined with complete
certainty ( 6 , p. 1053 ff.).

Perhaps the most imperative receptor need of the organism is
to have organs responsive to a state of injury within the internal
environment, i.e., to the destruction of tissue or to situations which
if intensified or long continued would result in the destruction of
tissue. The process of organic evolution has provided the higher
organisms with an abundant supply of such end organs. They are
sometimes called nociceptors. The hollow organs of the viscera
are provided with nociceptors which are activated by persistent
distention. The organs responsive to tissue injury, if any, also
have not yet been determined with certainty ( 6 , p. 1065 ff.).

In order that the organism may behave differentially to good
and bad food, it is clear that receptors yielding differential neural
reactions to such substances must be provided. Nature has solved
this problem by evolving chemical receptors for liquids — the gusta-
tory receptor organs, or taste buds, located mostly along the sides
and at the back of the tongue. The location of these end organs
in the mouth is highly adaptive. They will not be activated by

STIMULUS RECEPTION-ORGANISM SURVIVAL

sapid substances unless the latter are about to be ingested, yet the
process of ingestion will not have gone so far at the time of end
organ activation that rejection will not be easily possible ( 6 ,
p. 1005 ff.).

The smell or olfactory receptors are activated by certain gases
Quite appropriately the olfactory end organs are found in an upper
chamber of the nose where they will be in contact with eddies from
a constantly changing sample of the aerial environment of the
organism, incidental to the breathing activities ( 6 , p. 992). The
olfactory receptor is in a favorable position for guarding the ali-
mentary canal against undesirable substances and for guiding to
suitable food substances. Since gases characteristic of a substance
are apt to surround it for some distance, the receptor’s response to
these may evoke rejection before the substance even enters the
mouth (£, p. 992 ff.).

THE RECEPTION OF MOVEMENT

An exceedingly important sense mediating one aspect of the
internal environment is that known as proprioception, or kinaes-
thesis. The receptors of this sense organ lie mainly in the muscles
and joint capsules. Since they are stimulated by movements of
either the muscles or the joints, these receptors become of special
value in mediating all forms of activity involving a considerable
degree of muscular coordination ( 6 , p. 1072 If.). Ultimately we
shall see reason to believe that this seemingly humble and obscure
receptor plays an indispensable role in the most complex and subtle
activities ever evolved by nature — those of symbolic behavior or
thought.

In the continued survey of movement reception we find that
nature has evolved a remarkable organ which gives differential
and characteristic receptor response to angular or progressive mo-
tions of the head, whether the motion be active or passive. The
ultimate receptors lie within the labyrinth (6, p. 204 ff.). This
structure consists in the main of three small semi-circular canals
placed at right angles to each other within the head and intimately
related topologically to the internal ear. When the head is turned
through an angle or moved progressively, the receptor organs within
the labyrinth are stimulated in a distinctive manner, in this way
initiating characteristic neural impulses. The particular receptor
organs influenced by a given turning movement of the head will

PRINCIPLES OF BEHAVIOR

3 6

depend largely upon the position in which the head is supported
by the neck muscles. Thus adaptive reactions to angular or
progressive movements of the body as a whole must be based on
the combination or pattern (see p. 44 ff. and p. 349 ff.) of the
neural impulses arising in the labyrinth and those arising in the
proprioceptive end organs in the neck and other parts of the body.
It is accordingly not accidental that the nerve fibers extending to
the brain from both the labyrinth and the muscles largely enter
the same general portion of the brain, namely, the cerebellum.

THE RECEPTION OF THE SPATIAL. RELATIONSHIPS

Even a cursory observation of organisms shows that their sur-
vival depends upon their behavior being coordinated not only to
the objects and substances in their environment, but to the dis-
tances and directions of these from the organism and from each
other. We have already noted an exceedingly crude distance re-
ceptor in hairs, some of which are specialized for this function.
Moreover, the proprioceptive and labyrinthine receptors are clearly
connected with active and passive movement, and so are closely
related to distance and direction.

The olfactory receptor must also be mentioned in this connec-
tion. Certain bodies, substances, and organisms give off gases, and
these gases spontaneously diffuse through the atmosphere. In gen-
eral the concentration of these gases will be greater, and so the
intensity of receptor response will be the more intense, the closer
the gas-emitting object is to the receptor. Intensity of olfactory
receptor response accordingly becomes a basis for the mediation
to the organism of a spatial relationship. The fact that gases are
disseminated by means of air currents as well as by conduction has
both advantages and disadvantages from the point of view of
organism adaptation. If the air current reaches the receptor after
passing near the redolent object, activation will occur, but if not,
the gas will never reach the receptor and so the olfactory end organ
will fail as a distance receptor.

A much more dependable distance receptor has been developed
incidental to the evolution of a receptor responsive to vibrations
in the air. The ultimate end organs of this receptor lie in the
cochlea of the inner ear. The reason the sense of hearing is a
good distance receptor is that any vibrating object sets the neigh-
boring air into vibration at the same rate, and this vibration is

STIMULUS RECEPTION-ORGANISM SURVIVAL 37

propagated to adjacent air with gradually diminishing amplitude
at the rate of a little over a thousand feet per second. This means
that if an auditory receptor sensitive to a frequency of 100 vibra-
tions per second begins sending neural impulses into the nervous
system, an object vibrating at the rate of 100 per second must be
somewhere in the external environment.

The differential response of the auditory receptors to the par-
ticular distance and direction of the vibrating object is mediated
in a striking manner. Other things equal, the intensity of the
vibratory impact on the end organ varies inversely with the dis-
tance. Unfortunately, this is a somewhat ambiguous criterion of
distance, since the intensity of vibratory impact is also dependent
upon the vibrational amplitude of the originating object.

The matter of the receptor response to direction is more com-
plex. This seems to be jointly dependent upon two factors. Since
there is an ear at each side of the head, the ear which is turned
toward the origin of an auditory vibration receives a somewhat
more intense vibratory impact than does the other ear. A second
factor, also dependent upon the double nature of the auditory re-
ceptor, is that the phases of the air waves vary a little at the
respective ears when one ear is turned toward the source as com-
pared with the identity of the phase occurring when both ears are
equally near the source. Moreover, the extent and nature of this
dissimilarity in wave phase vary continuously as the head is
rotated through 180 degrees from a position in which one ear is
turned directly toward the vibration source.

The queen of the senses and the distance and direction receptor
par excellence is the eye. The ultimate end organs of vision are
microscopic rods and cones imbedded in the retina. The accessory
mechanisms of the eye such as the iris, the lens, and the various
humors are merely means for bringing the light energy reflected
from the environment into adequate contact with the retina.

Owing to the action of the lens system of the eye, limited por-
tions of the external environment are projected with considerable
fidelity upon the retina. In this way the spatial pattern of light
frequencies reflected from environmental objects is presented di-
rectly to the retina, giving precise and characteristic neural respon-
siveness corresponding to each element of the pattern. This in-
comparable sense organ, with its responsiveness to the different
intensities of the various wave lengths of light, thus furnishes the

PRINCIPLES OF BEHAVIOR

receptor basis for an almost unlimited degree of differential adap-
tive reaction to distinct stimulus objects and situations.

Passing now to the matter of visual distance reception proper,
we find that the physics of light furnishes a reliable foundation in
the fact that the image of any given object as projected upon the
retina varies inversely in size with the distance of the object from
the eye. There is a slight ambiguity here in that the size of the
image is also dependent upon the size of the object itself, which may
vary considerably.

On the retina directly back of the pupil is a point of especially
clear vision called the fovea. Other things equal, detailed sensi-
tivity to visual patterns grows progressively less with distance from
the fovea toward the periphery of the retina. Now it happens that
the eye, unlike the ear, is a very mobile organ. It thus comes about
that organisms readily learn to roll the eye in its socket in such
a way that the image of an object of importance for adaptive re-
action will fall on the fovea of each eye. This is called fixation.
The movements of each eye in its socket are produced by the
action of a set of six small external muscles. It is clear that the
tension of the muscles which turn the eye in its socket, in conjunc-
tion with the tension of the muscles of the neck, must yield a com-
bination or pattern of proprioceptive neural impulses unambigu-
ously correlated with the direction of the object fixated.

Just as the doubleness of the hearing organ facilitates the re-
ception of auditory distance cues, so the fact that we have two
eyes aids in the reception of visual distance cues. Since the eyes
are some inches apart, fixation on a single point near at hand
produces a certain amount of convergence, and the closer the ob-
ject, the greater the convergence. This may easily be verified by
asking a friend to look at your finger as you move it forward and
backward from six to eighteen inches before his face. As the eyes
turn inward, the tension of the muscles performing this action
varies inversely with the distance of the object fixated. The pro-
prioceptive receptors in certain of these muscles are accordingly
stimulated to an extent which varies inversely with the distance
of the object. In this roundabout way, involving the joint action
of vision and proprioception, the differential sensory reception of
the distance of objects within from 60 to 100 feet is accomplished
with considerable precision.

STIMULUS RECEPTION-ORGANISM SURVIVAL

THE MEDIATION" OF TEMPORAL RELATIONSHIPS

Careful observation of the conditions to which organisms must
adapt themselves if they are to survive shows that in addition to
the qualitative and spatial characteristics of the environment, the
timing of behavior is frequently very important. The processes
of organic evolution appear to have solved this problem in an even
more roundabout and obscure manner than some of the receptor
problems hitherto considered. In fact, the temporal characteris-
tics of environmental events seem to be mediated without the
action of any special receptor organ at all.

It will be shown later (Figures 3 and 4) that the frequency
of neural impulses emitted by a stimulated receptor undergoes a
characteristic change during the continued action of the unchanged
stimulating energy. There is also reason to believe that the effects
of a stimulation which has ceased, persist for some time (p. 41),
meanwhile undergoing progressive and consistent diminution.
These changes in the neural responses to stimulation, almost purely
as a function of time, are believed to furnish organisms with an
adequate basis for timing their movements both during the con-
tinuance of a critical stimulus and after its termination.

THE PRIMARY PRINCIPLE OF STIMULATION

The first step in the neural mediation of the state of both the
internal and the external environment to the effectors of the or-
ganism is dependent upon the principle of stimulation or excita-
tion. The critical characteristic of this principle is that a small
amount of energy acting on some specialized structure will release
into activity potential energy from some other source, often in
relatively large amounts. A familiar example is the trigger action
of a gun. The projectile is impelled by the energy stored in the
explosive charge; the pressure of the finger on the trigger merely
serves to initiate the explosive action which, once started, is self-
propagated.

The principle of stimulation is operative at several points in
the integrative apparatus of the mammalian organism, two of
which we need to consider here. The first is the action of an
energy source, such as light, upon a receptor organ such as the
eye, which initiates a self-propagating impulse in an afferent nerve
fiber. The second is the action of an efferent neural impulse when

4°

PRINCIPLES OF BEHAVIOR

it impinges upon a muscle fiber. In this second case there results
a release of energy stored in the cell which takes the form of
longitudinal contraction ( 2 ).

QUALITATIVE VERSUS QUANTITATIVE RECEPTOR ANALYSIS

OF ENVIRONMENTAL ENERGIES

Modem neurophysiological studies have shown that the neural
discharges initiated by the stimulation of all receptors are substan-
tially alike — a series of discrete waves (Figure 3). This seems to

Fia. 3. Reproduction of a photographic record of the action potentials
from a single optic-nerve fiber of a horseshoe crab when the end organ was
stimulated with different intensities of light. The gaps in the records represent
the passage of 2.8, 1.4, 4.5, and 3.3 seconds respectively. In record A, stimula-
tion was .1 light unit; in B, .01; in C, .001; and in D, .0001. The duration of
the action of the light on the end organ is indicated by the shadowy line just
above the time line in each record. Note that the stronger the stimulation,
the more rapid the neural impulses; and the longer the duration of the light,
the slower the impulse emission. Note, also, that the light acts for an ap-
preciable time before any neural impulses are emitted (latency) ; that this
latency is shorter, the more intense the light energy; and that usually there
are a few neural impulses emitted after the termination of the light. This is
known as the after-discharge. (After Graham, 6, p. 830.)

indicate that the only means whereby qualitative differentiation of
environmental energies can take place in the nervous system must
lie in the differentiation of the nerve fibers which transmit the
impulses. Such an arrangement clearly provides a basis whereby
the automatic switchboard activities of the central nervous system
may route the impulses initiated by qualitatively distinct stimuli
(distinct forms of energy) to different muscles and muscle com-
binations.

STIMULUS RECEPTION-ORGANISM SURVIVAL

4 1

But if behavior is to be thoroughly adaptive, it must vary with
the quantitative differences in environmental energies as well as
with their qualitative differences. Observation indicates that
higher organisms do, in fact, react differentially to varying intensi-
ties of exactly the same forms of activating energy. If the radio
gives forth too weak a tone we turn the volume knob in one direc-
tion, and if it gives forth too strong a tone we turn the knob in
the opposite direction. Delicate physiological investigations have
revealed that the frequency of the neural waves or impulses emitted
by a receptor is slow with weak stimulation and fast with strong
stimulation. This principle is illustrated very nicely by a series of
records published by Graham (6) and reproduced as Figure 3. The
frequency of afferent neural impulses is accordingly the code re-
ceived by the central nervous system which differentiates the
various intensities of the same environmental energy. Somehow
the central nervous system is evidently able to route nerve currents
of different frequencies to the several muscle groups in much the
same manner that it does neural impulses coming in over different
fiber paths.

CHARACTERISTICS OF THE AFFERENT NEURAL IMPULSE

AND ITS PERSEVERATION

It is clear that the immediate determinant of action in organ-
isms is not the stimulating energy, but the neural impulse as finally
routed to the muscles. A presumably critical neural determinant
intermediate between these two extremes of stimulus (5) and re-
sponse (72) is the afferent neural impulse (s) at about the time it
enters the central ganglia of the nervous system. It is important
to note that this afferent impulse (s) varies in certain ways not
paralleled exactly by changes in the stimulating energy. A par-
ticular form of this lack of parallelism is shown clearly in Adrian’s
graph reproduced as Figure 4 and may be seen more concretely in
Figure 3. In all receptors the frequency of receptor discharge be-
gins at a low value and rapidly rises to a comparatively high
maximum, after which the rate gradually falls, even though the
stimulus continues to act without change.

Certain molar behavioral observations render it extremely
probable that the after-effects of receptor stimulation continue to
reverberate in the nervous system for a period measurable in
seconds, and even minutes, after the termination of the action

4 2 PRINCIPLES OF BEHAVIOR

of the stimulus upon the receptor. Apparently the receptor after-
discharge (see Figure 3) is much too brief to account for the
observed phenomena. Lorente de No (6, p. 182) has demon-
strated histologically the existence of nerve-cell organizations
which might conceivably serve as a locus for a continuous circular
self-excitation process in the nerve tissue. Rosenblueth (8) has

Fio. 4. Graph showing the rise and gradual fall in the frequency of im-
pulses emitted by the eye of an eel during continuous stimulation by a light
of 830-meter candles. Note that the maximum response of the end organ is
not reached until between .3 and .4 of a second after the beginning of stimula-
tion. (Adapted from Adrian, 1, p. 116.)

presented experimental evidence indicating that neural after-dis-
charges affecting the heart rate may persist for several minutes.
This is shown graphically in Figure 5. Such bits of evidence as
these tend to substantiate the stimulus-trace hypothesis of Pavlov
(7, pp. 39-40) which is utilized extensively in the present work.
Indeed, were it not for the presumptive presence of stimulus traces
it would be impossible to account for whole sections of well-authen-
ticated molar behavior, notably those involving the adaptive timing
of action.

THE HYPOTHESIS OF NEURAL INTERACTION

We must now note a second plausible neural hypothesis which
is also related to the view that the afferent impulse set in motion

STIMULUS RECEPTION-ORGANISM SURVIVAL

by a particular objective stimulus element is not the same under
all circumstances. It would appear that a given afferent receptor
discharge (s) is modified by interaction with other receptor dis-
charges (s 2 or S3) on their way to or actually entering the central
nervous system at about the same time. For example, a small patch
of gray paper (s t ) resting on a large piece of blue paper (s 2 ) will
be reported by a subject as yellowish, but when resting on a large
piece of red paper (s 3 ) it will be reported as greenish. It seems
probable that the neural impulses initiated by the light rays arising

Fiq. 5. Graphic representation of the reflex slowing down of the heart
rate of a cat and its recovery after cessation of the stimulus. Left vagus and
depressor stimulated centrally; right vagus cut; accelerators intact. Ordinates:
slowing of the heart per 10 seconds. Stimulation was at the rate of 7.8 per
second during the period indicated by the heavy portion of the base line.
(Adapted from Rosenblueth, 8 , Figure 6A, p. 303).

from the gray patch interact with the afferent impulses arising from
adjacent portions of the retina in such a way as to change some-
what the course of each. An afferent neural impulse which has
been modified by interaction is represented by the sign s. In the
case of vision, ample anatomical possibilities for such interaction
exist in lateral connections among afferent fibers in the end organ
itself, in nuclei lying between the end organ and the brain, and
in the projection centers of the brain where the afferent impulses
are received.

In the case of afferent impulses entering the brain from distinct

PRINCIPLES OF BEHAVIOR

types of receptors such as those for light and sound, there is less
opportunity for neural interaction than within a given receptor
such as the retina. But still the richness of interconnections in
the brain furnishes an ample basis for a certain amount of neural
interaction here also. This presumption of a tendency for afferent
neural impulses to interact before evoking organismic behavior has
not yet received detailed physiological proof, and it accordingly
has the status of an hypothesis. It will hereafter be called the
hypothesis of afferent neural interaction, or, more briefly, the neural-
interaction hypothesis.

The neural-interaction hypothesis becomes of special importance
when it is understood that organisms usually must behave in such
a way as to survive in situations which, from the stimulus point of
view, are exceedingly complex in the sense of involving the activa-
tion of large numbers of receptors at the same time. Moreover, to
be adaptive the behavior must sometimes occur only in the presence
of a particular combination or configuration of afferent receptor
impulses and not in response to any one of the component impulses;
sometimes the situation is reversed — the reaction must occur in
response to certain component impulses but not to the whole
compound; and, finally, the response must sometimes be made to
the component impulses irrespective of the occurrence or non-
occurrence of other receptor impulses. It is obvious that such a
seemingly inconsistent state of the environment presents an ex-
tremely difficult problem to the reacting organism, though not neces-
sarily so to the theorist who would explain the degree of success in
adaptation which various organisms actually manifest. It will be
shown later (p. 349) that the hypothesis of afferent neural inter-
action when combined with three other behavioral principles will
enable us to understand how organisms are able to react to patterns
or configurations of stimulation as such to the extent that this exists
in fact. A notable situation of this kind is the response of or-
ganisms to distance as received through binocular vision (p. 38 ff.).

THE SPONTANEOUS EMISSION OF IMPULSES BY NERVE CELLS

In this connection we must note another basic neural phenom-
enon which seems to have rather far-reaching effects upon molar
behavior. This is the apparent capacity of individual neurons
spontaneously to generate neural impulses, as contrasted with the
long-recognized capacity of nerve cells to continue the propagation

STIMULUS RECEPTION-ORGANISM SURVIVAL

of a neural impulse transmitted to them from some other source.
Many observations, such as that of the striking variability in or-
ganismic behavior under seemingly constant external conditions,
have suggested some such tendency. Paul Weiss has demonstrated
the phenomenon by means of an extremely clever experiment:

Fragments of spinal cord, including several segments, excised from
larval salamanders . . . were grafted into the gelatinous connective tissue
of the dorsal fin fold. ... In seven of the fourteen animals thus operated
a limb was grafted at some distance anteriorly or posteriorly to the cord
graft. . . . Histological study revealed . . . outgrowth of bundles of nerve
fibers into the surroundings. The outgrowing nerve fibers form connec-
tions with skin, trunk muscles, and in the presence of a grafted limb, also
with the latter. . . . Within a few weeks of the transplantation these
isolated cord limb complexes begin to exhibit functional activity, in which
three successive phases can be roughly identified. . . . The first phase is
characterized by intermittent or almost incessant twitching of the limb
muscles. The twitches usually appear in spells, starting with irregular
fibrillations and gradually building up to violent convulsions. ... At the
peak of activity, the contractions are remarkably well synchronized, the
limb executing strong periodic beats, sometimes at fairly regular intervals
of the order of one to several seconds. ... As a crucial check against the
possible intrusion of host innervation . . . the portion of the back con-
taining the grafted units was completely excised and tested in isolation.
Even so, the preparations exhibited the same functional activities as before.
{9, p. 350-352.)

It is at once evident that the spontaneous firing of nerve cells,
if general throughout the nervous systems of normal adult organ-
isms, must when taken in conjunction with the neural-interaction
hypothesis imply an incessantly varying modification of both
afferent and efferent impulses. From the point of view of the latter
it is to be expected that this would produce both qualitative and
quantitative variability of reaction to identical environmental re-
ceptor stimulation, this variability presumably being a function of
the normal “law” of probability. This variability of response is
called oscillation (p. 304 ff.) .

SUMMARY

For most animals, activity is necessary for survival. But not
just any movement is sufficient ; to be adaptive, movement must be
coordinated in a precise manner with the state of various portions
of the total environment. This coordination can occur only if the
state of the environment is somehow continuously brought to bear

4 6

PRINCIPLES OF BEHAVIOR

on the motor apparatus. Organic evolution has provided animals
with special organs for receiving these physical environmental
energies and converting them into neural impulses ; with a central
mass of neural tissue which, acting as a kind of automatic switch-
board, does a fair job of routing the receptor impulses in the direc-
tion of the several muscles in adaptive amounts and proportions;
and with efferent fibers for transmitting these routed impulses to
the individual muscular elements, thus evoking the adaptive
behavior.

Receptor responses are prime examples of stimulation — the re-
lease of relatively large amounts of resident energy by the action
of small amounts of energy from an external source. Reception of
touch, temperature, and pain is comparatively simple and requires
no comment. Articulatory movement is received by special pro-
prioceptive end organs. Progressive and angular movement is
received by the vestibular receptor. Spatial relationships are ob-
tained in various indirect ways, notably through the patterning of
impulses received through the ears and eyes. Temporal relation-
ships are also received in various indirect ways but chiefly through
the progressive diminution both in the frequency of impulse emis-
sion by the receptor during stimulation and in the strength of the
“stimulus trace” following the termination of stimulation.

The reactions of organisms are ultimately evoked by the neural
impulses which are relayed by the central nervous system to the
muscles and glands. In a very real but indirect sense, however,
reaction is evoked by the stimulus energies dependent on stimulus
objects, though the neural impulse arising from the stimulation of
a given receptor is by no means constant, even during the steady
action of such a stimulus. The stimulus trace is a hypothetical
perseverative process in the receptor areas of the brain which is
believed to follow the termination of a receptor discharge with
gradually decreasing strength for some seconds and possibly
minutes.

Another factor preventing complete agreement or “constancy”
between the stimulation element and the afferent discharge to the
brain is believed to be neural interaction — the mutual modification
of all impulses active in the nervous system at any given time, but
especially afferent impulses, more particularly those arising in the
same compound receptor organ. This hypothesis makes possible
the explanation of many important behavior phenomena, otherwise
inexplicable, such as the power of organisms to react to patterns

STIMULUS RECEPTION-ORGANISM SURVIVAL

of stimulation as distinguished from the elements making up the
pattern. Finally there must be added the presumptive phenom-
enon of spontaneous periodic emission of neural impulses by all
the cells of the nervous system. This, coupled with the interaction
hypothesis, explains many obscure behavior phenomena, notably
the variability or oscillation (4, p. 74) of behavior under what may
be presumed to be relatively static environmental conditions.

In the light of the preceding considerations, we formulate the
following as primary molar behavior principles, postulates, or laws :

POSTULATE 1

When a stimulus energy (S) impinges on a suitable receptor organ, an
afferent neural impulse (s) is generated and is propagated along con-
nected fibrous branches of nerve cells in the general direction of the
effector organs, via the brain. During the continued action of the
stimulus energy (S), this afferent impulse (s), after a short latency, rises
quickly to a maximum of intensity, following which it gradually falls to
a relatively low value as a simple decay function of the maximum. After
the termination of the action of the stimulus energy (S) on the receptor,
the afferent impulse (s) continues its activity in the central nervous
tissue for some seconds, gradually diminishing to zero as a simple decay
function of its value at the time the stimulus energy (S) ceases to act.

POSTULATE 2

All afferent neural impulses (s) active in the nervous system at any
given instant, interact with each other in such a way as to change each
into something partially different (s) in a manner which varies with every
concurrent associated afferent impulse or combination of such impulses.
Other things equal, the magnitude of the interaction effect of one afferent
impulse upon a second is an increasing monotonic function of the magni-
tude of the first.

NOTES

Pavlov’s Statement of the Principle of Afferent Neural Interaction

While Pavlov did not make much systematic use of the principle of afferent
neural interaction, he stated it clearly and explicitly, as is shown by the following
quotations assembled by Woodbury (10). The italics in all cases are Woodbury’s,
In connection with the phenomenon of conditioned inhibition, Pavlov remarked :
“When the additional stimulus or its fresh trace left in the hemispheres coincides
with the action of the positive stimulus, there must result some sort of special
physiological fusion of the effect of the stimuli into one compound excitation partly
differing from and partly resembling the positive one” (7, p. 70)

Later, when dealing with what in the present work are called stimulus patterns

PRINCIPLES OF BEHAVIOR

(p. 349 ff.), Pavlov stated: “The cases mentioned above show that a definite inter-
action takes place between different cells of the cortex , resulting in a fusion or syn-
thesis of their physiological activities on simultaneous excitation. (7, p. 144)

“Plainly the experiments reveal the great importance of the synthesizing
activity of the cortical cells which are undergoing excitation. These cells must
form, under the conditions of a given experiment, a very complicated excitatory
unit, which is functionally identical with the simple excitatory units existing in
the case of more elementary conditioned reflexes. Such active cortical cells must
necessarily influence one another and interact with one another , as has clearly been
demonstrated in the case of compound simultaneous stimuli. The mutual inter-
action between the excited or inhibited cortical elements in the case of compound
successive stimuli is more complicated; the effect of an active cortical cell upon
the one next excited varies according to the influence to which it was itself sub-
jected by the cell last stimulated. In this way it is seen that the order in which a
given group of stimuli taking part in a stimulatory compound are arranged, and
the pauses between them are the factors which determine the final result of the
stimulation, and therefore most probably the form of the reaction. . . .” (7,
pp. 147-148)

-.While Pavlov clearly recognized the principle which we have called afferent
neural interaction as well as the importance of its r61e in the process of condi-
tioning reactions to compound stimuli in distinctive combinations or patterns,
it is noteworthy that he did not recognize how the conditioning of reactions to
stimulus patterns as such can be derived. In effect this means that he left the
process of patterning as a primary principle. It can, however, be derived as a
secondary phenomenon from four of his other principles (p. 349 ff.) which appear
to be true primary molar laws :

1. Afferent neural interaction (as indicated by the above quotations)

2. Experimental extinction (p. 258 ff.)

3. The generalization of excitation effects (irradiation of excitation, p. 183)

4. The generalization of extinction effects (irradiation of inhibition, p. 262)

Afferent Neural Interaction and the Configuration Psychologies

There is reason to believe that most of the Gestalt writers make extensive but
implicit use of a principle which is substantially equivalent in some respects at
least to the principle of afferent neural interaction. Kohler, however, has been
explicit in this respect. In connection with a discussion of perceptual conscious
states, he remarks in a recent publication: “Our present knowledge of human
perception leaves no doubt as to the general form of any theory which is to do
justice to such knowledge: a theory of perception must be a field theory. By this
we mean that the neural functions and processes with which the perceptual facts
are associated in each case are located in a continuous medium; and that the
events in one part of this medium influence the events in other regions in a way
that depends directly on the properties of both in their relation to each other.”
(5, p. 55)

Again, in the same general context, Kohler states even more explicitly : “To
the extent, therefore, to which these observations bear witness to an interaction
which cannot as such be observed within the phenomenal realm, they cannot be
understood in purely psychological terms. According to our general program

STIMULUS RECEPTION-ORGANISM SURVIVAL 49

we shall therefore assume that the interaction occurs among the brain correlates

of the perceptual facts in question.” (5, p. 63)

Continuing the elaboration of this same general point of view, Kohler adds:
“If in a certain sense the correlate of a percept may be said to have a circum-
scribed local existence we shall none the less postulate that as a dynamic agent
it extends into the surrounding tissue, and that by this extension its presence is
represented beyond its circumscribed locus. There is no contradiction in these
statements. So far as certain properties of the percept process are concerned,
this process may be confined within a restricted area, and with this nucleus the
percept itself may be associated as an experience. At the same time the presence
of such a percept nucleus may lead to further events in its environment, of which
we are for the most part not directly aware ; but this halo or field of the percept
process may be responsible for any influence winch the process exerts upon other
percept processes.” (5, p. 66)

It seems likely that in case an attempt were made to utilize Kohler’s inter-
action principle in developing a thoroughgoing theory of the reaction of organisms
to stimulus configurations, it would need to be supplemented by additional
principles analogous to those required to supplement Pavlov’s parallel principle,
and for the same reasons.

REFERENCES

1. Adrian, E. D. The basis of sensation. New York: W. W. Norton and

Co., 1928. # _ ,

2. Fulton, J. F. Muscular contraction and the reflex control of movement .

Baltimore: Williams and Wilkins Co., 1926.

3. Hull, C. L. The problem of stimulus equivalence in behavior theory.

Psychol. Rev., 1939, 46, 9-30.

4. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. Mathematico-deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

5. Kohler, W. Dynamics in -psychology. New York, Liveright Pub. Co.,

1940.

6. Murchison, C. A handbook of general experimental psychology. Worces-

ter, Mass.: Clark Univ. Press, 1934.

7. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). Oxford Univ.

Press, 1927.

8. Rosenblueth, A. Central excitation and inhibition in reflex changes of

heart rate. Amer. J. Physiol., 1934, 107, 293-304.

9. Weiss, P. Functional properties of isolated spinal cord grafts in larval

amphibians. Proc. Soc. Exper. Biology and Medicine, 1940, 44, 350-352.

10. Woodbury, C. B. The learning of stimulus patterns by dogs. J. Comp.
Psychol. 1943, 36, 29-40..

CHAPTER IV

The Biological Problem of Action and Its Coordination

The receptors of an organism may respond with neural impulses
in code corresponding to the near-by presence of food, of an enemy,
or of a potential mate. But for the food to be seized and digested,
the enemy to be escaped, or the process of reproduction to be
initiated, the organism must do something; i.e., it must act. Just
as in the last chapter we surveyed the manner in which the processes
of organic evolution have solved the receptor problem, so in the
present one we shall consider how nature has evolved a solution
to the problem of action.

The effector activity of higher organisms is of two major kinds
— secretional and motor. Generally speaking, the control of adap-
tive secretion, such as that of saliva, seems to follow the molar laws
of movement. Indeed, some of the most important molar laws
of learning were originally isolated by Pavlov and his pupils
( 2 , p. 19 ff.) through the study of conditioned salivary secretions.
Because of their greater variety and general interest, discussion in
the present chapter will be confined largely to the motor effectors.

THE MOTOR ORGAN

In contrast to the diversity of organs evolved for the reception
of environmental energies, the equipment evolved for the execution
of movement is comparatively without variety. It consists of only
one type of organ — the muscle. There is plenty of variety in the
behavior of organisms, but the variety arises mostly from the
location and attachments of the several muscles and the permuta-
tions and combinations of their joint action rather than from their
essential structure. From the point of view of behavioral adapta-
tion, the characteristic function of muscle is contraction. By “con-
traction” is meant longitudinal shortening which, of course, neces-
sarily means transverse thickening.

The microscopic structure of muscle parallels to a considerable
extent its gross structure. The muscle cells are elongated bodies,
relatively thick in the middle and tapering at the end. These fibers
are activated in their contraction by neural impulses flowing in

THE BIOLOGICAL PROBLEM OF ACTION 5 1

from the central nervous system along the fibrous branches of nerve
cells. The actual junction of the nerve fibers with the individual

Fia. 6. A section in profile of the motor end-plate of a striated muscle,
from a young mouse. The vertical striations at the bottom of the figure
represent the muscular tissue. (From Fulton, after Boeke, 1 , p. 198.)

muscle cell is made by a specialized structure called the neural
end-plate . Typical connections between nerve fibers and muscles
are represented in Figure 6.

THE ALL-OR-NONE LAW OF MUSCLE-FIBER ACTION

Just as the intensity of stimulation needs to be transmitted to
the organism by means of a graded neural code, so the rate and
general vigor of muscular contraction need to be regulated and
controlled in order that adaptation may be adequate. Movement
in some situations, such as the flight of a mouse when pursued by
a cat, must be rapid and even violent in intensity, whereas in
others, such as that of the cat in stalking the mouse, it must be
slow and gentle. Nature has evolved a solution to the problem
of the gradation of the intensity of action in the “all-or-none”
mode of muscle-fiber response to neural discharge.

Within recent years ingenious and delicate physiological inves-
tigations have succeeded in isolating small numbers of muscle fibers,
in systematically varying the intensity of the neural impulses dis-
charging into them, and in measuring the extent of the resulting

52 PRINCIPLES OF BEHAVIOR

muscular action. The outcome of such an experiment is repro-
duced as Figure 7. There it may be seen that as the intensity of
the neural discharge gradually rises, the amplitude of the reaction
of the muscle as a whole increases, not gradually but by a series
of sharply marked steps; again, as the intensity of the stimulus
gradually subsides, the step-wise behavior of the muscle is re-
versed. Microscopic observation of the individual muscle cells
under such experimental conditions confirms the hypothesis that the
step-wise rise and fall in the magnitude of muscular activity of

Fia. 7. Reproduction of a photographic record (above) of the movements
of a muscle made up of a very small number of fibers. Below is a graphic
record of the variation in the magnitude of electrical stimulation (break
shocks). Especially note, at the left of the record, that the magnitude of the
shock rises and falls gradually yet the muscular reaction rises and falls by a
series of discrete steps. These steps are believed to be due to the entrance
or cessation of the action of discrete muscle fibers, which is the substance of
the all-or-none law, viz., that a particular muscle fiber either responds
maximally or not at all, regardless of the amount of stimulation delivered.
(From Fulton, 1 , p. 52.)

Figure 7 is due to the fact that at each step in the reaction record
a new fiber of the muscle has become active or inactive respec-
tively. Meanwhile those fibers previously innervated by the weaker
neural discharge continue also to be innervated by all stronger
discharges. The conclusion from this and much other evidence
of a similar nature is that each individual muscle fiber has a reac-
tion threshold of its own, below which it will not respond at all
and above which it will respond with its maximum contraction.
Moreover, the several muscle fibers differ considerably in this
reaction threshold, i.e., in the intensity of neural discharge required

THE BIOLOGICAL PROBLEM OF ACTION 53

to activate them. Thus as the intensity of the neural discharge
into a muscle increases, the reaction thresholds of larger and
larger numbers of muscle fibers are passed, which brings more and
more of them into action. Since the contraction of the several
fibers of a muscle summates in their joint pulling action on their
tendinous attachment, it comes about that there is a gradation in
the intensity of muscular reaction from very slow and gentle reac-
tions to very rapid and violent ones. In large muscles consisting
of very many fibers, the step-wise action of Figure 8 is blurred
and obscured into the appearance of a smooth, composite, con-
tractile effect.

UNLEARNED COORDINATION OF MUSCULAR ACTIVITY

For some organisms, muscles alone suffice for locomotion and
other biological needs. Thus an earthworm is able to crawl about
and to secure food; if stimulated by a touch it can withdraw into
the safety of its burrow with quite remarkable suddenness. But
for the more exacting survival requirements of higher organisms,
muscles must be combined with relatively rigid levers. In lower
forms of life such as insects, the lever system is on the outside of
the body and serves also as a kind of protective armor. The levers
of higher organisms consist of bony structures within the body,
those levers primarily concerned with locomotion being generally
rod-like in form. In order that the bony levers may be moved
in various directions, muscles are usually attached to them in
pairs called antagonists. Thus the contraction of the biceps on
the top of the arm bends or flexes the arm at the elbow, whereas
the contraction of the triceps on the opposite side of the arm re-
verses the movement, straightening or extending the arm. By the
combined action of various bones, joints, and muscular contrac-
tions, terminal portions of the body such as the hand, say, may
be moved in almost any conceivable direction or combination of
directions.

The joint action of two or more muscles in the adaptive move-
ment of a portion of an animal’s body is called muscular coordina-
tion. Sometimes a dozen or more distinct muscles are involved in
a single coordinated movement, such as the extension of the hind
leg of the cat (Figure 8). This coordination (or integration, as it
is sometimes called) is brought about in the main by the action
of the nervous system. Some of these coordinations are so simple

PRINCIPLES OF BEHAVIOR

and uniform in their nature that relatively fixed and unchanging
modes of neural action have been evolved to perform them. In
experimentally simplified animal preparations it has been shown,
for example, that if a receptor discharge evokes the contraction of
one of two antagonistic muscles, the other member of the pair
relaxes; i.e., the second muscle automatically receives a neural dis-
charge which is inhibitory rather than excitatory in nature.

Certain coordinated movements are so universally required for
survival and the conditions under which they occur are so uniform

Fio. 8. Figure illustrating the muscles engaged in contraction during the
flexion-phase (A) and the extension phase (B) of the reflex step of the cat.
About a dozen different muscles are more or less active in each phase of the
stepping act. (From Fulton, 1, p. 448.)

that automatic neural mechanisms are provided at birth for their
execution. If a dog’s spinal cord is severed in the region between
the front and rear legs, the hind legs will drag when the animal
attempts to walk. But if the dog’s body is suspended in a kind
of hammock so that the legs hang free, and the toes of one foot
are pinched, the leg on which the toes are pinched will flex, i.e.,
draw upward, and at the same time the opposite leg will extend
itself. This is known as the stepping reflex. It will be noted that
if the intact dog were standing and one foot were lifted to escape
an injurious stimulus, the opposite leg would need to be extended

THE BIOLOGICAL PROBLEM OF ACTION 55

to support that end of the body; otherwise it would fall to the
ground.

Alternate flexion and extension of the legs is, of course, an
obvious characteristic of the locomotor process of the higher organ-
isms. In this connection it is illuminating to observe that if a
dog whose cerebrum or forebrain has been removed,

is lifted from the ground and supported in such a way that its feet hang
free, reflex walking, running, or galloping can be very easily elicited, and
it often commences spontaneously as soon as the limbs are allowed to
hang pendant under their own weight. If quiescent and the pad of one
hind foot is pinched, walking may commence, being initiated by a flexor
contraction on the side stimulated and an extensor contraction on the
opposite side ... If the stimulation is very intense the animal may
actually gallop, (I, p. 467)

REFLEX CHAINS INVOLVING INTERACTION WITH

THE ENVIRONMENT

Large numbers of other obviously adaptive movements are inte-
grated in this manner. A cinder in the eye will evoke lid closure
together with activity of the tear gland, the combined action usu-
ally resulting in the removal of the stimulus object. Bright moving
objects evoke eye movements resulting in ocular fixation, i.e., in
the turning of the eyes in their sockets so that the image of the
object will fall on the fovea.

A touch of the finger on the cheek of an infant, near its mouth,
will usually evoke a quick turning movement of the head toward
the side touched, and a simultaneous opening of the mouth. The
biological function of this coordinated movement combination is
evidently the seizure of the nipple. The head-turning movement
would be likely to result in the nipple entering the open mouth.
The contact of the nipple (or any similar object) with the cutane-
ous receptors in the mouth normally evokes complex coordinated
sucking movements. These in turn (in case the touching object
is the nipple) normally produce a flow of milk into the mouth.
The milk coming in contact with the gustatory and cutaneous re-
ceptors evokes salivary secretions and complex coordinated swal-
lowing movements which place the food in the upper portion of
the esophagus. This initiates peristalsis which deposits the food
in the stomach. The presence of the food in the stomach initiates
the complex secretional and motor processes of digestion, which
finally result in absorption, thus normally insuring nutrition and

PRINCIPLES OF BEHAVIOR

5 6

the survival of the organism in so far as it is dependent upon food.
It is noteworthy that one indispensable link in this chain — the
flow of the milk — depends upon the very specialized action of the
external environment, and once the milk is in the mouth the food
itself constitutes the medium which joins the several links of the
reflex chain.

SUMMARY

Generally speaking, higher organisms must be more or less
active or perish. The effector organs of the typical higher organ-
ism consist of glands and muscles. Movement is produced by the
longitudinal contraction of the individual fibers making up the
muscles. The several fibers of a muscle vary widely in the inten-
sity of the neural impulse necessary to evoke action. For all inten-
sities of neural impulse below the threshold the fiber will be wholly
inactive, and for all intensities above the threshold it will be equally
and maximally contractile.

The infinitely complex results produced by the simple muscular
contractions of organisms are brought about by the various com-
binations of a relatively small number of muscles, all contracting
various degrees and at various rates, several contractions jointly
determining the position of a bodily part such as a finger or a foot.

Certain adaptive situations are of such regularity that ready-
made chains of reflex receptor-effector connections are adequate
for survival, e.g., the blinking of the eyelid at any rough contact
with the cornea. In many chains of reflex activity the action of
the environment supplies indispensable links of the chain, as in the
suckling of a young animal.

REFERENCES

1. Fulton, J. F. Muscular contraction and the reflex control of movement.

Baltimore: Williams and Wilkins, 1926.

2. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

CHAPTER V

Characteristics of Innate Behavior Under Conditions

of Need

We saw in an earlier chapter (p. 17) that when a condition
arises for which action on the part of the organism is a prereq-
uisite to optimum probability of survival of either the individual
or the species, a state of need is said to exist. Since a need, either
actual or potential, usually precedes and accompanies the action
of an organism, the need is often said to motivate or drive the
associated activity. Because of this motivational characteristic of
needs they are regarded as producing primary animal drives.

DRIVES ARE TYPICAL INTERVENING VARIABLES

It is important to note in this connection that the general con-
cept of drive ( D ) 1 tends strongly to have the systematic status
of an intervening variable or X (see Figure 1), never directly ob-
servable. The need of food, ordinarily called hunger, produces a
typical primary drive. Like all satisfactory intervening variables,
the presence and the amount of the hunger drive are susceptible
of a double determination on the basis of correlated events which
are themselves directly observable. Specifically, the amount of
the food need clearly increases with the number of hours elapsed
since the last intake of food; here the amount of hunger drive (D)
is a function of observable antecedent conditions, i.e., of the need
which is measured by the number of hours of food privation. On
the other hand, the amount of energy which will be expended by
the organism in the securing of food varies largely with the inten-
sity of the hunger drive existent at the time; here the amount of
“hunger” is a function of observable events which are its conse-
quence. As usual with unobservables, the determination of the
exact quantitative functional relationship of the intervening vari-
able to both the antecedent and the consequent conditions presents

1 In caae the reader subsequently fails to recall the meaning of this, or
any of the other signs employed in the present volume, the significance may
be recovered in a moment by consulting the alphabetical list of signs and
their meanings given in the Glossary of Symbols (p. 403 ff.).

PRINCIPLES OF BEHAVIOR

serious practical difficulties. This probably explains the paradox
that despite the almost universal use of the concepts of need and
drive, this characteristic functional relationship is not yet deter-
mined for any need, though some preliminary work has been done
in an attempt to determine it for hunger (f).

INNATE BEHAVIOR TENDENCIES VARY ABOUT A CENTRAL RANGE

With our background of organic evolution we must believe that
the behavior of newborn organisms is the result of unlearned, i.e.,
inherited, neural connections between receptors and effectors ( 9 Ur )
which have been selected from fortuitous variations or mutations
throughout the long history of the species. Since selection in this
process has been on the intensely pragmatic basis of survival in
a life-and-death struggle with multitudes of factors in a consider-
able variety of environments, it is to be expected that the innate
or reflex behavior of young organisms will, upon the whole, be rea-
sonably well adapted to the modal stimulating situations in which
it occurs.

It may once have been supposed by some students of animal
behavior, e.g., by Pavlov and other Russian reflexologists, that
innate or reflex behavior is a rigid and unvarying neural con-
nection between a single receptor discharge and the contraction of
a particular muscle or muscle group. Whatever may have been the
views held in the past, the facts of molar behavior, as well as the
general dynamics of behavioral adaptation, now make it very
clear not only that inherited behavior tendencies (bUr) are not
strictly uniform and invariable, but that rigidly uniform reflex
behavior would not be nearly so effective in terms of survival in
a highly variable and unpredictable environment as would a be-
havior tendency. By this expression is meant behavior which will
vary over a certain range, the frequency of occurrence at that seg-
ment of the range most likely to be adaptive being greatest, and
the frequency at those segments of the range least likely to be
adaptive being, upon the whole, correspondingly rare. Thus in the
expression bUr, R represents not a single act but a considerable
range of more or less alternative reaction potentialities.

The neurophysiological mechanism whereby the type of flexible
receptor- effector dynamic relationship could operate is by no means
wholly clear, but a number of factors predisposing to variability
of reaction are evident. First must be mentioned the spontaneous

INNATE BEHAVIOR AND CONDITIONS OF NEED

impulse discharge of individual nerve cells, discussed above (p.
44). This, in conjunction with the principle of neural interaction
operating on efferent neural impulses ( efferent neural interaction ) ,
would produce a certain amount of variability in any reaction.
Secondly, the variable proprioceptive stimulation arising from the
already varying reaction would, by afferent neural interaction,
clearly increase the range of variability in the reaction. Finally,
as the primary exciting (drive) stimulus increases in intensity, it
is to be expected that the effector impulses will rise above the
thresholds of wider and wider ranges of effectors until practically
the entire effector system may be activated.

Consider the situation resulting from a foreign object entering
the eye. If the object is very small the stimulation of its presence
may result in little more than a slightly increased frequency of
lid closure and a small increase in lachrymal secretion, two effector
processes presenting no very conspicuous range of variability except
quantitatively. But if the object be relatively large and rough,
and if the stimulation continues after the first vigorous blinks and
tear secretions have occurred, the muscles of the arm will move
the hand to the point of stimulation and a considerable variety of
manipulative movements will follow, all more or less likely to
contribute to the removal of the acutely stimulating object but
none of them 'precisely adapted to that end.

In the case of a healthy human infant, which is hungry or is
being pricked by a pin, we have the same general picture, though
the details naturally will differ to a certain extent. If the need be
acute, the child will scream loudly, opening its mouth very wide
and closing its eyes; both legs will kick vigorously in rhythmic
alternation, and the arms will flail about in a variety of motions
which have, however, a general focus at the mouth and eyes. In
cases of severe and somewhat protracted injurious stimulation the
back may be arched and practically the entire musculature of the
organism may be thrown into more or less violent activity.

SOME PRIMARY NEEDS AND THE MODAL REACTIONS TO THEM

The major primary needs or drives are so ubiquitous that they
require little more than to be mentioned. They include the need
for foods of various sorts (hunger}, the need for water (thirst), the
need for air, the need to avoid tissue injury (pain), the need to
maintain an optimal temperature, the need to defecate, the need

PRINCIPLES OF BEHAVIOR

to micturate, the need for rest (after protracted exertion), the need
for sleep (after protracted wakefulness), and the need for activity
(after protracted inaction). The drives concerned with the main-
tenance of the species are those w r hich lead to sexual intercourse
and the need represented by nest building and care of the young.

The primary core or mode of the range of innate or reflex ten-
dencies to action must naturally vary from one need to another if
the behavior is to be adaptive. In cases where the role of chance
as to what movements will be adaptive is relatively small, the
behavior tendency may be relatively simple and constant. For
example, the acute need for oxygen may normally be satisfied (ter-
minated) by inspiration; the need represented by pressure in the
urinary bladder is normally terminated by micturition. It is not
accidental that these relatively stereotyped and invariable reactions
are apt to concern mainly those portions of the external environ-
ment which are highly constant and, especially, the internal en-
vironment which is characteristically constant and predictable.

In the case of mechanical tissue injury, withdrawal of the in-
jured part from the point where the injury began is the character-
istic reflex form of behavior, and the probability of the effectiveness
of such action is obvious. Environmental temperatures consider-
ably below the optimum for the organism tend to evoke shivering
and a posture presenting a minimum of surface exposed to heat loss.
Temperatures above the optimum tend to produce a general inac-
tivity, a posture yielding a maximum surface for heat radiation,
and rapid panting. In certain relatively complex situations such
as those associated with the need for food, water, or reproduction,
the factor of search is apt to be included as a preliminary. Since
extensive search involves locomotion, the preliminary activities
arising from these three needs will naturally be much alike.

ORGANIC CONDITIONS WHICH INITIATE THREE TYPICAL

PRIMARY DRIVE BEHAVIORS

During recent years physiologists and students of behavior have
made important advances in unraveling the more immediate con-
ditions which are associated with the onset of the activities char-
acteristic of the three most complex primary drives — thirst, hunger,
and sex. Thirst activities appear from these studies to be initiated
by a dryness in the mouth and throat caused by the lack of saliva,
which in its turn is caused by the lack of available water in the

INNATE BEHAVIOR AND CONDITIONS OF NEED 6 1

blood. The hunger drive seems to be precipitated, at least in part,
by a rhythmic and, in extreme cases, more or less protracted con-
traction of the stomach and adjacent portions of the digestive tract
presumably caused by the lack of certain nutritional elements in
the blood. Copulatory and maternal drives appear to be most
complex of all and are not too well understood as yet. It is known
that female copulatory receptivity (oestrum) is precipitated by the
presence in the blood of a specific hormone secreted periodically,
and that male copulatory activity is dependent upon the presence
in the blood of a male hormone. Just how these hormones bring
about the actual motivation is not yet entirely clear.

Because of the inherent interest of some of these studies and
of their presumptive aid in enabling the reader to orient himself
in this important phase of behavior, three or four typical investiga-
tions will be described.

TYPICAL STUDIES OF HUNGER- MOTIVATED ACTIVITY

The first study to be considered concerns hunger; it was per-
formed by Wada ( 8 ). This investigator trained human subjects
to swallow a tube with a small balloon at its end, the latter enter-
ing the stomach and the other end of the tube projecting from the

— t

">-•“* * •• tym.J

.-.-r

. t m

,j. • \

'"'••i, ,,tii

i } i/i 1 * “ 1 1

iii! liHU I JP 1 1 ! I : /

Fia. 9. A record of the restless movements of a sleeping student (middle
line) and the parallel (hunger) contractions of the student’s stomach. Note
that the sleeper’s restless movements coincide, in general, with the periods
of maximal stomach contraction. (After Wada, 3 , p. 29.)

subject’s mouth. Then the balloon was inflated and the free end
of the tube was attached to a delicate pneumatic recording mecha-
nism. The subjects slept through the night upon an experimental
bed which permitted the automatic recording of any restless move-
ments of the sleeper. A presumably typical record so obtained
is reproduced as Figure 9. The lower tracing of this record shows
at the extreme left the rise of a series of rhythmic contractions of

PRINCIPLES OF BEHAVIOR

the stomach, terminating in a kind of cramp followed by a period
of cessation. Presently the stomach contractions begin again, and
are more or less continuous throughout the remainder of the record.
But the main point of this is that the restless movements of the
sleeping student (recorded as short vertical oscillations of the
middle line in Figure 9) occurred as a rule only when the stomach
contractions were occurring , especially when they were at a maxi-
mum.

Richter (£) attempted to secure parallel records of the stomach
contractions and the restless locomotor activity of rats and other
organisms to complete the proof of the presumptive relationship
afforded by Wada’s findings, but was unsuccessful, apparently be-
cause of technical difficulties encountered. However, he was able

A -\ 1 1 1 1 1

Time «n hours

Stomach contractions

Fia. 10. Diagrammatic representation of the inferred relationship between
the periodic stomach contractions of rats and their restless locomotor activity
in the living cage. (After Richter, 2 , p. 312.)

to show that rats are periodically rather restless in the living cage
for a short time before going into a food chamber to eat, after
which general activity quickly subsides. The periodicity of these
restless movements is about the same as that known to occur with
the stomach contractions. Richter accordingly concludes from a
convincing array of such indirect evidence that the relationship
between random, restless activity and the gastric hunger contrac-
tions is substantially that shown in Figure 10. It is to be observed
that this figure is not a record but, rather, a diagrammatic repre-
sentation of an inferential relationship. Nevertheless, Richter’s
reasoning is fairly convincing and Figure 10 quite probably repre-
sents the true situation.

The functional interpretation of this restless behavior is that
an organism which moves about more or less continuously will in

INNATE BEHAVIOR AND CONDITIONS OF NEED 63

general traverse a wide area and consequently will be more likely
to encounter food than if it remains quietly in one place.

TYPICAL STUDIES OF SEXUALLY MOTIVATED ACTIVITY

Richter has also shown that the male rat displays much more
restless locomotor activity when the sex drive is operating than
when it is not. He placed male rats in drum-like cages pivoted
on a central axis in such a way that if the animal attempted to
climb the circular side of the cage its weight would turn the drum.
Automatic counting devices aggregated the amount of this kind of

Fia. 11. Graphic representation of restless locomotor activity of a male
albino rat in a revolving cage before and after castration. (After Richter,
2 , 329.)

locomotor activity by days. A graphic representation of the nu-
merical values so obtained is shown at the left of Figure 11. At
about the 195th day the rat was castrated. Note the abrupt drop
in the restless locomotor activity. Even if one makes a certain
allowance for the shock produced by the operation as such, the
inference is that when the hormone secreted by the testes is in the
blood, the animal is generally active, but when this hormone is
withdrawn through castration, generalized locomotor activity falls
to a relatively low level and remains there.

Wang (4) has shown by analogous means that the female rat
is maximally active in this same restless fashion about every fourth
day, the two or three days between showing a relatively small
amount of activity. A considerable number of cycles from a single

PRINCIPLES OF BEHAVIOR

<*4

female rat are represented graphically in Figure 12 ( 2 ). That
these maxima of locomotor activity are coincident with periodically
recurring sex drive is shown by the fact that on the occasion of
the maxima such animals are receptive to the sexual advances of
the male.

The functional interpretation of these studies is similar to that

14,000,.

1 2,0001

«D

o 10.000.

4.000,

spool

Age \ ri davjs ^

70 8b 9b 10b lib l£b 13 b 146

Fia. 12. Graphic representation of the locomotor activity of a female
albino rat over a series of days. Note the prevalence of a four-day cycle of
activity. On the days of maximum activity these animals are sexually
receptive. (After Richter, 2 , 321.)

of the investigations involving hunger; an organism which moves
about continuously will traverse a wide area and consequently
will be more likely to encounter a mate than will an organism
which remains in a single place.

SUMMARY

Animals may almost be regarded as aggregations of needs. The
function of the effector apparatus is to mediate the satiation of

INNATE BEHAVIOR AND CONDITIONS OF NEED 65

these needs. They arise through progressive changes within the
organism or through the injurious impact of the external environ-
ment. The function of one group of receptors (the drive recep-
tors) is to transmit to the motor apparatus, via the brain, activat-
ing impulses corresponding to the nature and intensity of the need
as it arises. Probably through the action of these drive receptors
and receptor-effector connections preestablished by the processes
of organic evolution, the various needs evoke actions which increase
in intensity and variety as the need becomes more acute.

Because of the inherently fortuitous nature of the environmental
circumstances surrounding an organism when a state of need arises,
the kind of behavior which will be required to alleviate the need
is apt to be highly varied. For this reason rigid receptor-effector
connections could not be very effective in terms of organismic sur-
vival. Accordingly we find, as a matter of fact, that innate molar
response to a given state of need presents a considerable range of
activity, the activity often consisting of a sequence of short cycles
of somewhat similar yet more or less varied movements. Such
behavior cycles are believed normally to show a frequency dis-
tribution in which those acts most likely to relieve the need occur
most often, and those acts less likely to terminate it occur corre-
spondingly less often. Thus reflex organization ( a U R ) has more
than one string to its bow ; if one reaction cycle does not terminate
the need, another may. The modal form of reaction will also be
strongest, so that it will usually occur not only most frequently but
earliest and probably will remedy the situation ; but if the environ-
ment chances to be such that some other simple action sequence is
required, in due course this action sequence will probably occur
and the organism will survive. Finally, in still more complicated
situations a particular combination of these acts may terminate the
need. In this way innate behavior tendencies are organized on a
genuine but primitive trial-and-error basis.

Lastly, it is to be noted that just as food-seeking activity
begins long before the organism is in acute need of food, so other
drives become active long before heat or cold or pain becomes seri-
ously injurious or even in the least harmful. In short, it may be
said that drives become active in situations which, if more intense
or if prolonged, would become injurious. Once more, then, the
probability aspect of primitive behavior tendencies becomes mani-
fest.

PRINCIPLES OF BEHAVIOR

In the light of the preceding considerations we formulate the
following primary molar behavior principle:

POSTULATE 3

Organisms at birth possess receptor effector connections ( sUr ) which,
under combined stimulation (S) and drive (D), have the potentiality of
evoking a hierarchy of responses that either individually or in combi-
nation are more likely to terminate the need than would be a random
selection from the reaction potentials resulting from other stimulus and
drive combinations.

NOTES

The Role of Adaptation in Systematic Behavior Theory

The emphasis in this and preceding chapters on the general significance of
organic evolution in adapting organisms to meet critical biological emergencies
calls for a word of comment, lest the reader be misled in regard to the rdle that
adaptation, as such, plays in the present system. It is the view of the author
that adaptive considerations are useful in making a preliminary survey in the
search for postulates, but that once the postulates have been selected they must
stand on their own feet. This means that once chosen, postulates or principles
of behavior must be able to yield deductions in agreement with observed detailed
phenomena of behavior; and, failing this, that no amount of a 'priori general
adaptive plausibility will save such a postulate from being abandoned.

Problems Associated with the Use of Drive ( D ) as an Intervening Variable

Most writers on behavior theory utilize the concept of need or some equivalent
such as drive, though hardly one of them has faced squarely the associated problem
of finding the two equations necessarily involved if the concept of drive is to take
its place in a strict mathematical theory of behavior (5). In the case of hunger,
for example, there must be an equation expressing the degree of drive or motiva-
tion as a function of the number of hours’ food privation, say, and there must
be a second equation expressing the vigor of organismic action as a function of
the degree of drive (D) or motivation, combined! in some manner with habit
strength. A correlated task of some magnitude is that of objectively defining a

unit in which to express the degree of such a motivatio nal intervening variable
(see p. 226 ff.).

Now it is a relatively easy matter to find a single empirical equation expressing
vigor of reaction as a function of the number of hours’ food privation or the
strength of an electric shock, but it is an exceedingly difficult task to break such
an equation up into the two really meaningful component equations involving
hunger drive (Z)) or motivation as an intervening variable. It may confidently
be predicted that many writers with a positivistic or anti-theoretical inclination
will reject such a procedure as both futile and unsound. From the point of view
of systematic theory such a procedure, if successful, would present an immense
economy. This statement is made on the assumption that motivation ( D ) as
such, whether its origin be food privation, electric shock, or whatever, bears a

INNATE BEHAVIOR AND CONDITIONS OF NEED 67

certain constant relationship to action intensity in combination with other factors,
such as habit strength. If this fundamental relationship could be determined
once and for all, the necessity for its determination for each special drive could not
then exist, and so much useless labor would be avoided. Unfortunately it may
turn out that what we now call drive and motivation will prove to be so hetero-
geneous that no single equation can represent the motivational potentiality of any
two needs. Whether or not this is the case can be determined only by trial.

REFERENCES

1. Perin, C. T. Behavior potentiality as a joint function of the amount of

training and the degree of hunger at the time of extinction. J. Expet ;

Psychol., 1942, 30, 93-113.

2. Richter, C. P. Animal behavior and internal drives. Quar. Rev. Biology,

1927, t, 307-343. ,

3. Wada, T. An experimental study of hunger in its relation to activity.

Arch. Psychol., 1922, No. 57.

4. Wanq, G. H. The relation between “spontaneous” activity and oestrous

cycle in the white rat. Comp. Psychol. Monog., 1923, No. 6.

5. Young, P. T. Motivation of behavior. New York: John Wiley and Sons,

1936.

CHAPTER VI

The Acquisition of Receptor-Effector Connections

Primary Reinforcement

We have seen above that organisms require a considerable
variety of optimal conditions if the individual and the species are
to survive. In many cases where the conditions, particularly in-
ternal ones, deviate materially from the optimum, complex auto-
matic physiological processes make the adjustment. An example
of this is the remarkable manner in which the blood is maintained
at a practically constant state in the face of a great variety of
adverse conditions. This type of automaticity has been called by
Cannon the “wisdom of the body” (£)• In the case of certain
other needs, and here lies our chief interest, the situation is remedi-
able only by movement, i.e., muscular activity, on the part of the
organism concerned. The processes of organic evolution have pro-
duced a form of nervous system in the higher organisms which,
under the conditions of the several needs of this type, will evoke
without previous learning a considerable variety of movements
each of which has a certain probability of terminating the need.
This kind of activity we call behavior.

THE PROBLEM AND GENERAL NATURE OF LEARNING

It is evident, however, that such an arrangement of ready-made
(inherited) receptor-effector tendencies, even when those evoked
by each state of need are distinctly varied, will hardly be optimally
effective for the survival of organisms living in a complex, highly
variable, and consequently unpredictable environment. For the
optimal probability of survival of such organisms, inherited be-
havior tendencies must be supplemented by learning. That learning
does in fact greatly improve the adaptive quality of the behavior
of higher organisms is attested by the most casual observation.
But the detailed nature of the learning process is not revealed by
casual observation; this becomes evident only through the study
of many carefully designed and executed experiments.

The essential nature of the learning process may, however, be
stated quite simply. Just as the inherited equipment of reaction

PRIMARY REINFORCEMENT 6 9

tendencies consists of receptor-effector connections, so the process
of learning consists in the strengthening of certain of these con-
nections as contrasted with others, or in the setting up of quite
new connections. In many ways this is the highest and most sig-
nificant phenomenon produced by the processes of organic evolu-
tion. It will be our fascinating task in the present and several
succeeding chapters to tease out bit by bit from the results of very
many experiments the more important molar laws or rules accord-
ing to which this supremely important biological process takes

place.

In accordance with the objective approach outlined in Chapter
II we must regard the processes of learning as wholly automatic.
By this it is meant that the learning must result from the mere
interaction between the organism, including its equipment of action
tendencies at the moment, and its environment, internal as well
as external. Moreover, the molar laws or rules according to which
this interaction results in the formation or strengthening of recep-
tor-effector connections must be capable of clear and explicit state-
ment. Recourse cannot be had to any monitor, entelechy, mind,
or spirit hidden within the organism who will tell the nervous
system which receptor- effector connection to strengthen or which
receptor-effector combination to connect de novo. Such a pro-
cedure, however it may be disguised, merely raises the question of
the rule according to which the entelechy or spirit itself operates;
this, of course, is the original question all over again and clarifies
nothing.

THE STRENGTHENING OF INNATE RECEPTOR-EFFECTOR

CONNECTIONS

Because of its presumptive temporal priority in the life of the
organism, we shall consider first the problem of the selective
strengthening of one among a variety of inherited movement ten-
dencies evoked by a need in a particular environing situation. This
can perhaps best be done by means of an illustrative experiment,
even though some of the reaction tendencies there operative may
already have been modified by learning. The experimental pro-
cedure and the results will be described in a little detail, out of
consideration for readers who have slight knowledge of the routine
methodologies characteristic of behavior laboratories.

7 <>

PRINCIPLES OF BEHAVIOR

Demonstration Experiment A. The laboratory in which the experiment
is performed is without windows and its walls are painted black; this gives
the room an appearance of being rather dimly illuminated, though in fact
it is not. On a table rests a black wooden apparatus about two feet long,
a foot wide, and a foot high. It has a hinged glass lid which permits clear
observation of the interior. The floor of the box consists of small trans-
verse rods of stainless steel placed about a quarter inch apart. Midway
between the two ends of the box is a partition consisting of the same
type of metal rods similarly arranged but placed vertically. This partition
or barrier reaches to within about four inches of the lid. A two-throw
electric switch permits the charging of the floor rods of either compart-
ment and of the partition with a weak alternating current.

On a second table nearby there rests a wire cage containing a sleek
and lively albino rat about one hundred days of age. The laboratory
technician opens the lid of the cage and the rat at once stands up on
its hind legs with its head and forepaws outside the aperture. The tech-
nician grasps the rat about the middle with his bare hand and transfers
it to one of the compartments of the apparatus. The animal, after a
brief pause, begins moving about the compartment, sniffing and inspecting
the various parts, often stretching up on its hind legs to its full length
against the walls of the box.

After some minutes the technician throws the switch which charges
both the partition and the grid upon which the rat is standing. The
animal's behavior changes at once; in place of the deliberate exploratory
movements it now displays an exaggeratedly mincing mode of locomotion
about the compartment interspersed with occasional slight squeaks, biting
of the bars which are shocking its feet, defecation, urination, and leaps up
the walls. These reactions are repeated in various orders and in various
parts of the compartment; sometimes the same act occurs several times in
succession, sometimes not. After five or six minutes of this variable be-
havior one of the leaps carries the animal over the barrier upon the
uncharged grid of the second compartment. Here after an interval of
quiescence and heavy breathing the animal cautiously resumes exploratory
behavior, much as in the first compartment. Ten minutes after the first
leap of the barriei the second grid is charged and the animal goes through
substantially the same type of variable behavior as before. This finally
results in a second leaping of the barrier and ten minutes more of safety,
after which this grid is again charged, and so on. In this way the animal
is given fifteen trials, each terminated by a leap over the barrier.

A comparison of the animal’s behavior leading to his successive
escapes from the charged grid shows clear evidence of learning in
that upon the whole the time from the onset of the shock to the
escape became progressively less, until at the last few trials the
leaping reaction followed the onset of the shock almost instantane-
ously. Meanwhile the competing reactions gradually decreased in
number until at the end they ceased to occur altogether. Once or

PRIMARY REINFORCEMENT

7 *

twice the rat even leaped the barrier before the shock was turned
on at all. Here, then, we have a clear case of selective learning.

It is evident from the foregoing that the final successful com-
petition of the reaction of leaping the barrier (-R*) with the various
futile reactions of the series such as leaping against the wooden
walls of the apparatus {R t ) y squeaking ( Rg) y and biting the floor
bars (I2 3 ) must have resulted, in part at least, from a differential
strengthening of R+. It is also evident that each of these com-
peting reactions was originally evoked by the slightly injurious
effects of the current on the animal’s feet (the condition of need
or drive, D) in conjunction with the stimulation (visual, cutaneous,
etc.) arising from the apparatus at about the time that the reaction
took place. The stimulation arising from the apparatus at the
time of the respective reactions needs to be designated specifically:
leaping against the wall will be represented by squeaking, by
Sa' ; biting, by S A ” *, and leaping the barrier, by S A It is assumed
that preceding the learning, the leaping of the barrier was evoked
by a compound connection between the receptor discharges Sd and
s Ay arising from Sd and S A respectively, and R^ \ i.e., R * must have

been evoked jointly by the converging connections, sd >R± and

Sa „ > R k , These are the connections which evidently have been

strengthened or reinforced. Because of this, learning is said to be
a process of reinforcement .

We must now approach the central problem of learning by at-
tempting to formulate the rule according to which primary rein-
forcement occurred in this case of selective learning. More specifi-
cally, we must ask the rule according to which the connections

s D > R k and s A "’ » R h were differentially strengthened so as

to become dominant over the numerous other reaction tendencies.
The most plausible statement of this rule at present available is:
Whenever a reaction ( R ) takes place in temporal contiguity with
an afferent receptor impulse ($) resulting from the impact upon a
receptor of a stimulus energy (5), and this conjunction is followed
closely by the diminution in a need {and the associated diminution
in the drive, D, and in the drive receptor discharge, s D ), there will
result an increment, A (s >R) y in the tendency for that stim-

ulus on subsequent occasions to evoke that reaction. This is the
“law” of primary reinforcement }

Thus in the case of learning exhibited in Demonstration Ex-

1 Actually, of course, this formulation has only the status of an hypothesis.
The term law is here used in much the same loose way that Thorndike has

PRINCIPLES OF BEHAVIOR

7 *

periment A, both of the afferent stimulus impulses, s D and s A "',
were obviously active when reaction R k occurred because they
evoked it. Moreover, this conjunction of s D and s A — with R k was
followed immediately by the termination of the shock effects or
need, and so by a reduction in s©. But by the principle of primary
reinforcement just formulated this reduction in the need and drive

INTERIOR OF APPARATUS (A>~*-S A — >^-S A S A * S $ A ”*

Fio. 13. Diagrammatic representation of the process of strengthening or
reinforcing the connections between s© — » r« and s*'"— »r. The step-like rises
and falls of the several horizontal lines such as those of D and s D , the shock
to the tissue and Sd, represent the rise from zero and the fall of the respective
processes. The arrows with wavy shafts >) represent a physical causal

relationship other than by way of receptor-effector stimulus evocation. Thus
the rise of the current on the grid (Sd) causes the shock to the tissue of the
animal’s feet, the response of the receptor in the skin (sd) of those regions,
and the drive (D) or motivation to action. The separation of the foot from
the grid by the act of jumping ( R «) terminates simultaneously the injurious
action or need, the receptor discharge (s D ), and the drive (D), though the
current on the grid remains unchanged. It is the reduction in the drive
receptor impulse (s©) and the drive (D) which are believed to be the critical

factors in the process of reinforcement. The arrows with solid shafts ( »),

whether curved or straight, separate or jointed, represent receptor-effector
relationships in existence before the learning process here represented occurred.

The arrows with broken shafts ( >) represent receptor-effector connections

here in process of formation. Distance from left to right represents the
passage of time.

(D) will, through the associated decrement in the drive receptor im-
pulse ($z>), result in increments, A (s A "' »#*) and A ( s D >

Rk)) to the tendency for such conjoined afferent stimulus elements
(s A "' and Sn) to evoke the reaction (R+) on subsequent occasions.
The major dynamic factors of this process are represented diagram-
matically in Figure 13.

used it in his famous expression, “law of effect,” to which the above formu-
lation is closely related (9, p. 176).

PRIMARY REINFORCEMENT

THE ACQUISITION OF NEW RECEPTOR-EFFECTOR CONNECTIONS

We proceed now to the consideration of the formation of a
genuinely new receptor-effector connection. This turns out to be
only a special case of the law of primary reinforcement which we
have just formulated. Moreover, this type of selective learning
may be demonstrated by means of an experiment differing only
slightly from the one already considered at some length.

Demonstration Experiment B. The variation of Demonstration Ex-
periment A consists merely in the sounding of a buzzer continuously
from a time two seconds before the shock is turned on the grid until the
animal leaps the barrier. The course of the learning is much as in the
preceding experiment up to the point where the animal has eliminated
all of the original acts except that of leaping the barrier. At this point,
however, the animal begins occasionally to leap over the barrier during
the first two seconds of the sounding of the buzzer.

It will be evident at once that this outcome follows directly
from the law of primary reinforcement stated above because, as

Fio. 14. Diagrammatic representation of the dynamic factors involved in
the setting up of a new receptor-effector connection in a selective learning
situation, Demonstration Experiment B. With the exception of the upper
line representing the rise and continuation of the buzzer stimulus and its

receptor discharge, and the connection sb > R « in process of formation,

this diagram is exactly the same as that of Figure 13.

shown by Figure 14, the receptor discharge (s*) resulting from the
action of the buzzer vibrations (5*) on the mechanisms of the
internal ear has a conjunction (just as have s D and s.*'") with
and the termination of the need, and that of the associated drive
(D) and drive receptor discharge (sd), constitute a reinforcing

PRINCIPLES OF BEHAVIOR

state of affairs exactly as in the first form of the experiment. Thus
after reinforcement in the manner described there exists in the body
of the organism a habit represented diagrammatically by the

broken-shafted arrows of Figures 14 and
15.

From other experiments (p. 209) it is
known that the receptor-effector connec-
tions from several stimulus aggregates, all
converging simultaneously upon the same
reaction, tend strongly to summate (see
p. 223 ff.). As a result the action of all
three connections would be stronger than
that of the entire group less any one. Be-
cause of the absence of S D > s D > R+

during the uncharged state of the grid, it
is to be expected that the rat would at
first wait until the delivery of the shock
before jumping. However, with addi-
tional reinforcements the action-evocation
strengths of S B and S A ••• finally become
great enough when combined to evoke the
reaction before the shock is delivered.
Because this reaction is the same as that
which usually occurs only after the onset
of the shock, it is called antedating or
anticipatory; since it results in total escape from the injury pro-
duced by the shock, it is highly adaptive.

Finally, with still more reinforcements, the receptor-effector
connections become so strong that the relatively static connection,

Sa'" > R\, alone will evoke the jumping reaction, i.e*»

the stimulus of the apparatus alone will evoke the adaptive re-
sponse, before the onset of either the buzzer or the shock.

Fia. 15. Diagram repre-
senting the results of a re-
inforcement in which a
quite new receptor-effector

connection (sb * R*)

has been set up as shown
in Figure 14. The arrows
with double shafts repre-
sent the combination of
the old and the newly ac-
quired receptor - effector
connections between
and 8d, respectively, and
R<. The unbroken shafts
represent either unlearned
receptor-effector connec-
tions or at least connec-
tions in existence before
the particular learning un-
der consideration began.

THE CONDITIONED REFLEX

A special case of the action of the principle of reinforcement
sketched above is found in the type of experiment in which there
is set up what is indifferently called the conditioned reflex or the
conditi(med reaction . Despite a certain artificiality, the relative
simplicity of this type of experiment has permitted the isolation
of a large number of molar laws, particularly in the laboratory of

PRIMARY REINFORCEMENT

I. P. Pavlov, the great Russian physiologist (5). The typical
conditioned-reflex experiment is conducted in such a way as (1) to
set up new receptor-effector connections rather than, as in Demon-
stration Experiment A, to strengthen connections already strong
enough in combination to evoke overt reactions, and (2) to elimi-
nate the necessity of selecting one reaction from the numerous
varied reactions normally evoked by the conjunction of a need in
a stimulating situation. The elimination of the complication char-
acteristic of selective learning is brought about by the simple ex-
pedient of reinforcing the first, i.e., the dominant, reaction of the
potential sequence of competing tendencies normally evoked by a
need situation. Since the process of primary reinforcement em-
ployed in such experiments usually involves the termination of the
need, the duration of the need is so brief that the weaker members
of the potential action group rarely become overt.

In order to make the distinctive characteristics of the condi-
tioned-reflex experiment especially clear, the following example of
conditioned-reflex learning was designed in such a way as to be
an exact parallel to Demonstration Experiment A.

Demonstration Experiment C. A dog is habituated to stand in a stock
or wooden framework resting on a laboratory table. A soft leather
moccasin containing an exposed electric grid in its sole is laced to the dog s
foot. The moccasin is attached to a light hinged board which is held
down by a coil spring. The electric circuit which conducts the alternating
current to the grid passes across a connection which is broken when the
dog’s footboard is lifted one inch at the point of moccasin attachment.

The experiment is conducted as follows: A buzzer is sounded two
seconds before a shock is delivered to the dog’s foot through the moccasin
grid. The resulting shock produces as its dominant reaction a reflex
lifting of the shocked foot which breaks the circuit, thus terminating
the shock. No doubt the dog makes many other muscular contractions
in addition to those which result in the lifting of the foot, but these are
usually neglected in such experiments; the main point is that the foot-
lifting act always does take place at once. After ten minutes the buzzer-
shock combination is repeated with the same results as before. This is
continued until fifteen reinforcements have taken place.

By the principle of primary reinforcement outlined above, the
receptor discharge (s c ) 1 arising from the buzzer vibrations (the
“conditioned stimulus” or S c ) is temporally conjoined with the foot-

1 The symbols sc, su, and Ru have become conventionalized in the con-
ditioned-reflex literature and are here used with their conventional meaning.
The dot has been placed above the s in so in order also to conform to the
usage of the present work.

PRINCIPLES OF BEHAVIOR

7<*

lifting reaction (the so-called “unconditioned reaction” or R u ) f and
this conjunction is followed at once by the termination of the shock
or need and of the associated drive receptor impulse (s„) which
constitutes the reinforcing state of affairs. As a result there must
be set up an increment to the associative connection between the
afferent receptor impulse produced by the buzzer vibrations and
the foot-lifting reaction. Thus after a sufficient number of repeti-
tions there must arise the new superthreshold receptor-effector con-
nection, So > Ru, and the dog begins regularly to lift his foot

promptly at the sound of the buzzer.

Quite naturally, exactly as in the preceding experiments, there
is also set up a connection between the various static stimuli aris-
ing from the apparatus (S A ) and the reaction ( R u ) because the
former also are active in conjunction with the foot-lifting act and
so become connected along with the so-called conditioned stimulus,
thus: 54 > R u - As a result of this connection the dog will fre-

quently lift his foot when the buzzer is not sounding, just as the
rat would sometimes leap the barrier when neither the buzzer nor
the shock was acting. In the case of the dog this unadaptive be-
havior is discouraged by the spring which tends to hold down the
footboard.

THE CONDITIONED REFLEX A SPECIAL CASE OF ORDINARY

LEARNING REINFORCEMENT

Demonstration Experiment C presents a fairly typical example
of conditioned-reflex learning, though it is characteristic of the
school of Bechterev ( 1 ) rather than that of Pavlov. 1 We have
already seen that the acquisition of a quite new receptor-effector
connection, a phenomenon of conditioned-reflex learning, is deduc-
ible from the conditions of Demonstration Experiment C on the
basis of the law of primary reinforcement formulated above in con-
nection with a typical bit of selective learning (Demonstration
Experiment A). Because of the current differences of opinion con-
cerning the relationship between selective learning and conditioned-
reflex learning, an explicit and somewhat detailed comparison of
them as types will now be made. In order to facilitate such a

1 The salivary conditioned-reflex technique of Pavlov involves certain
complexities, the consideration of which must be delayed until the next chap-
ter when the necessary groundwork for their adequate understanding will
have been laid.

PRIMARY REINFORCEMENT 77

comparison, Figure 16 has been constructed to represent in con-
siderable detail the dynamic factors here conceived to be involved
in conditioned-reflex learning, in close parallel with the representa-

BUZZER (S c ) -
APPARATUS (S A )

DRIVE AND DRIVE
IMPULSE FROM

INJURY TO TISSUE

ELECTRIC CHARGE
MOCCASIN GRID

Fig. 16. The conditioned-reflex learning of Demonstration Experiment C
represented according to the primary reinforcement formulation presented
in the text. The R u terminates the shock or drive (So) which in turn termi-
nates both the tissue injury and the associated receptor impulses. Presumably
the latter is the essential constituent of the reinforcing state of affairs which
sets up the connections £o' * Ru and

tion of the process of selective reinforcement of Demonstration
Experiment A as presented in Figure 13.

A comparative study of Figures 13 and 16 verifies explicitly
what has already been pointed out. There it may be seen that:

1. Simple selective learning (Figure 13) involves the selection of a
particular reaction from numerous alternative reactions which are evoked
more or less at random by the need and the stimulus situation jointly,
whereas in conditioned-reaction learning (Figure 16) there is only one
(and the same) conspicuous act involved at each reinforcement trial.

2. Simple selective learning may involve the mere strengthening oi
receptor-effector connections, already of superthreshold strength before
the beginning of the experiment, whereas in conditioned-reflex learning
there results, typically, a completely new receptor-effector connection.

From this point of view, Demonstration Experiment B offers
a kind of transition from the extremes of Demonstration Experi-

iCEPTOR %
)CK (D.s u )

(NEED)

. \

s . — v-

n O

°R,. (LIFTING OF FOOT)

PRINCIPLES OF BEHAVIOR

ments A to C, since it is clearly a case of selective learning yet it
involves the setting up of a receptor-effector connection de novo
and, at the same time, the presumptive strengthening of other con-
nections already superthreshold in strength. It must be added
that conditioned-reaction experiments may also be arranged in
such a way as to strengthen superthreshold connections and that
the terminal phases of conditioned-reaction learning inevitably in-
volve the strengthening of connections set up de novo in the early
stages of a given learning process.

These last considerations suggest that the differences between
the two forms of learning are superficial in nature; i.e., that they
do not involve the action of fundamentally different principles or
laws, but only differences in the conditions under which the prin-
ciple operates ( 6 , Theorem I). This preliminary impression is
confirmed when we make a comparison of the two situations (as
represented by Figures 13 and 16). On one critical point both
cases are identical — the reinforcing state of affairs in each consists
in the abolition of the shock injury or need, together with the asso-
ciated decrement in the drive and drive receptor impulse, at once

after the temporal conjunction of the af-
ferent receptor discharge and the reaction.
This is, of course, all in exact conformity
with the law of primary reinforcement
formulated above (p. 71).

We pass now to the consideration of
an alternative interpretation of the con-
ditioned reflex, namely, that held by Pav-
lov (8), its greatest exponent. This is
represented fairly well by Figure 17, a
diagram frequently employed to illus-
trate Pavlov’s views by American writers
of elementary textbooks (8, p. 381; 7,
p. 245). As this diagram suggests, Pav-
lov agrees substantially with the “law
of reinforcement” formulated above, as well as with the “law of
effect” as formulated by Thorndike, in holding that the conditioned
stimulus S c (or its trace, s c ) must have approximate temporal con-
junction with the unconditioned reaction (#«) before the receptor-
effector connection can be established. Pavlov differs from the law
of reinforcement by regarding as the critical element of the rein-
forcing state of affairs the occurrence of S u , in this case the onset

Fia. 17. Conventional
diagram of the dynamics
of conditioned-reflex learn-
ing common in American
textbooks of elementary
psychology. The circle sur-
rounding Rc is to indicate
that the preexperimental
reaction of So is inconspic-
uous, unknown, or non-ex-
istent.

PRIMARY REINFORCEMENT 79

of the shock. On the other hand, the critical element in the rein-
forcing state of affairs by our own hypothesis is the reduction in
the drive receptor impulse (sd or s u ) which accompanies the reduc-
tion of the need, i.e., reduction of the physiological injury to the
tissue of the feet, caused by the termination of the shock.

It is an easy matter to show the inadequacy of Pavlov’s formu-
lation as a general theory of learning by applying it to the case
of simple selective learning presented by Demonstration Experi-
ment A (Figure 13). According to Pavlov’s hypothesis, every one
of the false reactions R I} R t , Rs, etc., should have been reinforced
just as should R^, because (in the conditioned-reflex terminology)
each is evoked by the unconditioned stimulus (S u )* In that case
if any selection at all could be expected to occur that reaction
which takes place the most frequently would be the one selected
rather than the reaction which actually would set in motion a
causal sequence leading to the termination of the need. Yet innu-
merable experiments show that, other things equal, the reaction
which is followed at once by a diminution in a primary need will
be selected, regardless of its original frequency of occurrence.

It is not difficult to understand how Pavlov could have made
such an error. His mistaken induction was presumably due in part
to the exceedingly limited type of experiment which he employed.
Within the range of his restricted procedures his formulation was
consistent with all of the facts and observed relationships. Had
he worked even a little with simple selective learning he would
doubtless have seen his error and corrected it. Actually he seems
not to have occupied himself greatly with the problem of the exact
nature of the reinforcing state of affairs ; he was interested mainly
in discovering the detailed characteristics of conditioned-reflex phe-
nomena by an intensive experimental attack. In this he was
eminently successful.

SUMMARY

The infinitely varied and unpredictable situations of need in
which the higher organisms find themselves make any form of
ready-made receptor-effector connections inadequate for optimal
probability of survival. This natural defect of inherited reaction
tendencies, however varied, is remedied by learning. Learning
turns out upon analysis to be either a case of the differential
strengthening of one from a number of more or less distinct reac-

PRINCIPLES OF BEHAVIOR

tions evoked by a situation of need, or the formation of receptor-
effector connections de novo; the first occurs typically in simple
selective learning and the second, in conditioned-reflex learning.
A mixed case is found in which new receptor-effector connections
are set up at the same time that selective learning is taking place.

An inductive comparison of these superficially rather divergent
forms of learning shows one common principle running through
them all. This we shall call the law of primary reinforcement. It
is as follows: Whenever an effector activity occurs in temporal
contiguity with the afferent impulse, or the perseverative trace of
such an impulse, resulting from the impact of a stimulus energy
upon a receptor, and this conjunction is closely associated in time
with the diminution in the receptor discharge characteristic of a
need, there will result an increment to the tendency for that stim-
ulus on subsequent occasions to evoke that reaction. From this
principle it is possible to derive both the differential receptor-
effector strengthening of simple selective learning and the acquisi-
tion of quite new receptor-effector connections, characteristic of
conditioned-reflex learning as well as of certain forms of selective
learning. Pavlov puts forward the alternative hypothesis that the
critical element in the reinforcing state of affairs is the occurrence
of the unconditioned stimulus. This formulation fits conditioned-
reflex phenomena but breaks down when applied to selective learn-
ing situations, a fact which shows it to be an inadequate inductive
generalization. Fortunately the inadequacy of this interpretational
detail of Pavlov’s work in no way detracts from the scientific value
of the great mass of empirical findings produced by his laboratory.

NOTES

Is the Reinforcing State of Affairs in Learning Necessarily the “Effect”

of the Act Being Reinforced?

It has already been suggested that the hypothesis as to the reinforcing state
of affairs adopted in the present work is distinctly related to that of Thorndike’s
“law of effect.” Thorndike seems to have coined this expression because the
state of affairs which has been found empirically to be necessary in order to pro-
duce differential reinforcements, as in Demonstration Experiments A, B, and C,
under ordinary circumstances comes literally as the effect of the reaction which is
reinforced. This cause-and-effect relationship is shown explicitly by means of
the wavy-shafted arrows leading from the act reinforced, e.g., R 4 in Figures 13
and 14, which patently causes the termination of the injurious action to the foot
tissue and of the receptor impulses, s©. A strictly parallel though slightly more
complex causal relationship is seen in Figure 16, where Ru terminates the current

PRIMARY REINFORCEMENT

8 1

on the grid by interrupting the circuit and this, in turn, terminates the shock
to the foot and at the same time brings to an end the receptor impulses (s u ) arising
from the current passing through the receptor organs buried in the tissue.

At first thought it might be supposed that since only consistent reinforcement
will set up stable habits, and since the reaction ( R ) when in a given situation
yielding 5 to the receptors will be followed consistently by a reinforcing state of
affairs only when there is a causal connection between the antecedent events and
those which follow, the “law of effect” would be established on a firm a priori
foundation. In point of fact, however, this rule breaks down in the Pavlovian
conditioned-reflex experiment where the salivary reaction of the dog can by no
stretch of the imagination be regarded as the cause of the receipt of food which
reduces the hunger and is commonly considered the reinforcing agent in this
experiment. This paradox is explained by observing that some common cause,
the food, produces first the salivation and, later, the reduction in the need
(hunger). Accordingly the reinforcing state of affairs is temporally related to
the reaction involved in the reinforcement in strict accordance with the law-of-
reinforcement formulation, but not as the effect of the reaction being reinforced.
A second presumptive exception to Thorndike’s formulation of the nature of the
reinforcement process is the conditioned kneejerk (20). The termination of the
receptor discharge from the slightly injurious blow on the patellar tendon occurs
because of the brief duration of the impact of the hammer, rather than because
of the occurrence of the kneejerk.

Despite these minor exceptions, Thorndike’s inductive generalization, as
represented by the expression, “law of effect,” is based upon a very penetrating
bit of scientific insight into the dynamics of adaptive situations in general. Never-
theless the exception is probably genuine and it has seemed best to employ in
the present work the slightly more appropriate though less colorful expression,
law of reinforcement.

What Is the Critical Factor in Primary Reinforcement?

In Figures 13, 14, and 16 it will have been observed that reductions in (1) the
need and (2) the receptor response to the need both follow as consequences of the
act involved in the reinforcement process, directly in Figures 13 and 14 and
indirectly in Figure 16. These considerations raise the question as to which of
the two is to be regarded as the critical reinforcing agent; this can be determined
only when some radical experiment is performed in which one of the two is elimi-
nated and the other remains active. The writer is aware of no critical evidence
of this kind. Until such becomes available the issue must remain uncertain.
Meanwhile, in the interest of definiteness, the alternative of reduction in drive-
receptor response is chosen for use in the present work as the more probable of
the two. Should critical evidence later prove this choice to be in error, a correc-
tion can be made. In the present stage of our ignorance regarding behavior
dynamics, "an error in either direction would not seem to have such far-reaching
systematic implications as to render correction unduly difficult.

The Effectiveness of Reinforcement and the Intensity of the Need Involved

in the Reinforcement

A recent study by Finan (4) tends to support the view that reduction of need
is a critical factor in the primary reinforcement process. This investigator

PRINCIPLES OF BEHAVIOR

trained groups of albino rats to secure pellets of food by depressing a small bar in a
Skinner-Ellson apparatus (see pp. 87, 268). Each group of animals received
the same number of reinforcements but in a different condition of food privation.
Two days later, after the food need had been equalized, all groups were extin-
guished. The median number of non-reinforced reactions required to produce
a constant degree of experimental extinction were as follows :

Hours of food privation during reinforcement 1 12 24 48

Median number reactions to produce extinction 25 57.5 40 41

While the above extinction values indicate that the relationship is not a simple
increasing function of the number of hours’ food privation at the time of rein-
forcement, they do show that for some hours after satiation there is a progressive
increase in the effectiveness of reinforcement; thus the two phenomena are shown
definitely to be connected.

The Onset, Versus the Termination, of Need-Receptor Impulse as the

Critical Primary Reinforcing Factor

The view has been put forward in the preceding pages that the termination of
need-receptor impulse is the critical factor in the primary reinforcement process.
Many students in this field, however, have held the view that reinforcement is
critically associated with the onset of the need or drive, as represented by the
physiological shock in Demonstration Experiment C.

The evidence from innumerable selective learning experiments, as typified by
Demonstration Experiments A and B, leaves little doubt as to the soundness of
the need-reduction generalization. This does not necessarily mean that the
need-onset hypothesis is false; there may be more than one mechanism of rein-
forcement. While such seeming improvidence in biological economy appears
somewhat opposed to the principle of parsimony, it is not without parallel in other
fields; most organisms possess more than one means of excretion, and some
organisms possess more than one independent means of reproduction. Such
general considerations merely pose the question and warn us of multiple possi-
bilities; they cannot be decisive.

Turning to experimental evidence now availablejwe find that selective learning
of the type shown in Demonstration Experiments A and B usually yields results
consistent only with the termination hypothesis. On the other hand, the results
from conditioned-reflex experiments, typified by Demonstration Experiment C,
are consistent with either hypothesis. This ambiguity probably arises from the
brief duration of the shock usual in such experiments ; the onset and termination
of the need occur so close together that it is difficult clearly to distinguish the
influence of each. For example, it is quite possible that the critical reinforcing
factor in the conditioned kneejerk experiment (f0) often cited in this connection
(5, p. 85) may be the termination of the receptor discharge resulting from the
usually rather severe blow on the patellar tendon, which is the unconditioned
stimulus used to evoke this reaction. The matter is complicated still further by
the intrusion of the generalized results of previous learning, especially where highly
sophisticated human subjects are employed in the investigations. Thus the only
critical evidence now available seems to favor the reduction or termination
hypothesis ; decision must probably come from carefully designed and executed

PRIMARY REINFORCEMENT

experiments perhaps involving surgical interference with portions of the nervous

^ EMhe interim we shall proceed on the positive assumption that the termination
of the need (or of its closely correlated receptor response) is a primary reinforcing
factor ; this hardly seems open to doubt. Even if the onset of the nee ^' or °
correlated receptor response, proves to have genuine reinforcing capaci y, the
dynamics of behavior are such that it would not have much adaptive value.

REFERENCES

1 Bechterev V. M. General principles of human reflexology (trans. by E,
and W. Murphy from the Russian of the 1928 ed.). New \ork: Inter-
national Publishers, 1932. w w xT orton

2. Cannon, W. B. The wisdom of the body. New \ork. W . W. Nor 0

3. DashiS; j!^ 2. Fundamentals of general psychology. New York: Hough-

4. Finan^^L. Quantitative studies in motivation. I. Strength of condi-

tioning in rats under varying degrees of hunger. J. Comp. Psycho .,

5. Hi^rof V^and Marquis, D. G. Conditioning and learning. New

York: D. Appleton-Century Co., Inc., 1940. , , w

6. Hull, C. L. Mind, mechanism, and adaptive behavior. Psychol. Rev.,

7. Murphy^G. A briefer general psychology. New York: Harper and

8. PavlovQ. P. Conditioned reflexes (trans. by G. V. Anrep). London:

9. Thorndike, ^ L. The fundamentals 0 } teaming. New York: Teachers

College, Columbia Univ., 1932.

10 Wendt, G. R. An analytical study of the conditioned kneejerk. Arch.
Psychol, 1930, 19, No. 123.

CHAPTER VII

The Acquisition of Receptor-Effector Connections —

Secondary Reinforcement

The sequences of learned behavior considered in the last chap-
ter were all very short when perfected, only two or three seconds
at the most being required for their execution. These examples
were chosen not because they were especially typical of mammalian
behavior in general, but because they were relatively simple and
so lent themselves readily to an introductory exposition of the
principles of learning. We must now explicitly recognize the fact,
confirmed by universal observation of the everyday behavior of
animals including ourselves, that a great deal of behavior takes
place in relatively protracted sequences in which primary reinforce-
ment normally occurs only after the final act. Evidence will be
presented in a subsequent chapter (p. 139 ff.) showing that rein-
forcement probably must follow a receptor-effector conjunction
( sCr ) within about twenty seconds if it is to have an appreciable
effect. Consequently direct or primary reinforcement, as such, is
inadequate to account for a very great deal of learning. Fortunately
an ingenious series of experiments performed in Pavlov’s laboratory
in Petrograd has yielded a principle which explains these more re-
mote reinforcements. This supplementary reinforcement principle
is called secondary reinforcement. In the present chapter we shall
consider the nature, origin, and elementary functioning of this
extremely important principle.

DEMONSTRATION OF THE EXISTENCE OF SECONDARY

REINFORCEMENT

Because the principle of secondary reinforcement was first iso-
lated from the results of conditioned-reflex experiments, we shall
begin our presentation with an illustrative example from Pavlov’s
laboratory. The experiment was performed by Dr. Frolov; the
only account of this experiment available to English readers is
that of Pavlov (7, p. 34), who unfortunately omitted many of the

SECONDARY REINFORCEMENT 8 5

details necessary to an introductory account of this type of experi-
mentation. For the benefit of those unfamiliar with the method-
ologies of conditioned-reflex laboratories, these accessory details
are here supplied from the accounts of other relevant experiments.

Dr. Frolov experimented with a dog, one of whose salivary
glands had been diverted surgically so that the saliva discharged
through a fistula in the side of the animal’s face instead of flowing
into its mouth. Suitable apparatus was provided for the precise
determination of the number of drops secreted within a given time
interval. When hungry this dog would be presented with the tick-
ing of a metronome for a minute or so, and after 30 seconds meat
powder would (presumably) be blown into its mouth; the powder
would then be eaten by the dog, a considerable quantity of saliva
evoked by the incidental gustatory stimulation and chewing activ-
ity at the same time flowing from the fistula. After numerous
reinforcements of this kind it was found that the metronome acting
alone for 30 seconds evoked 13.5 drops of saliva; this is an ordinary
or “first-order” conditioned reflex. The above account presents a
fairly typical picture of conditioned-reflex learning by the Pav-
lovian technique.

Next, a black square was presented in the dog’s line of vision
for the first time; no saliva flowed from the fistula during this
stimulation. Following this test the black square was held in front
of the dog for 10 seconds, and after an interval of 15 seconds the
metronome was sounded for 30 seconds, no food being given. The
tenth presentation of the black square (alone) lasted 25 seconds;
during this period 5.5 drops of saliva were secreted. This is an
example of a “higher-order” conditioned reflex.

The conditions of Frolov’s experiment show that the visual
stimulation resulting from the presentation of the black square
had in some way acquired from association with the metronome
stimulation the capacity to evoke the salivary secretion independ-
ently. Since the presentation and consumption of food were not
associated with the acquisition of the second conditioned reaction,
it is assumed that during the original conditioning process the
metronome had not only acquired the capacity to evoke the flow
of saliva but had also acquired the capacity itself to act as a rein-
forcing agent. The metronome is accordingly said to be a secondary
reinforcing agent. For analogous reasons the resulting receptor-

effector connection (black square > salivation) set up by this

means is said to be a second-order conditioned reaction.

PRINCIPLES OF BEHAVIOR

SOME PROBLEMS CONCERNING SECONDARY REINFORCEMENT

Frolov’s experiment demonstrates in an unambiguous manner
the genuineness of secondary reinforcement, a first-rate scientific
achievement. Unfortunately, even when considered together with
the other Russian experiments in this field it leaves unanswered
numerous questions concerning the conditions necessary and suf-
ficient for secondary reinforcement to occur.

1. The reaction conditioned to the black square was qualitatively the
same as, though weaker than, that previously conditioned to the metro-
nome. Is this typical of secondary reinforcement, i.e., is secondary rein-
forcement confined to the transfer of the same reaction from one stimulus
to another, or may any receptor-effector conjunction be connected by

secondary reinforcement?

2. In Frolov’s experiment the metronome purports to have served a
double function: (a) that of evoking the reaction (salivation) which was
secondarily conditioned; and (6) that of reinforcing the conjunction of
the salivation thus evoked and the receptor discharge produced by the
presentation of the black square. Is this apparent duplication of function
by the metronome genuine and, if so, is it a characteristic or necessary
part of the secondary reinforcement process?

3. The receptor-effector conjunction involved in the setting up of
this second-order conditioned reflex was associated temporally not only
with the stimulation of the ticking metronome, but also with the evoca-
tion of the reaction already conditioned to the latter. This leads us to
ask: Are both of these events, i.e., both the presentation of the metronome
stimulus and the evocation of the reaction conditioned to it at the time
it acquired the power of being a secondary reinforcing agent, necessary
for the secondary reinforcement to occur, or is only one of these events
necessary for secondary reinforcement, and if only one, which one?

These and a number of other questions of a somewhat similar
nature arising from the Russian experiments in this field must be
examined. Before doing this, however, one or two general remarks
may be made concerning the situation as a whole. To a certain
extent these questions arise because of the distinctly artificial
nature of the conditioned-reflex experimental procedure. While m
no way detracting from the scientific significance of these investi-
gations, their artificiality probably does in some cases interfere
with the recognition of their bearing on the adaptive dynamics of
ordinary life situations. Accordingly, while considering the prob-
lems in question, an attempt will be made gradually to place the
subject of secondary reinforcement in a more natural functional

perspective.

SECONDARY REINFORCEMENT

WHAT REACTIONS MAY BE SECONDARILY REINFORCED?

We shall begin with the first of the above questions— whether
the reaction involved in secondary reinforcement must necessarily
be the same as that already conditioned to the secondary reinforc-
ing stimulus, or whether any reaction whatever may be so connected.
In order to find an answer to this question it will be necessary to
consider some experiments which differ considerably from the t> pe
employed by Pavlov. The first of these has been reported by Skin-
ner (P, p. 82).

This investigator fed a hungry rat tiny pellets of specially pre-
pared food by activating a food magazine which dropped one pellet
into a food cup at each activation. On each occasion the action
of the magazine produced a clearly audible sound vibration. In
the course of such training rats soon learn to interrupt whatever
they are doing when the magazine vibration occurs, go directly to
the cup, and eat the pellet. After 60 pellets had been given in this
way, the food was removed from the magazine and the rat left to
itself.

A horizontal brass bar projected from the wall of the apparatus
several centimeters above the food cup. In its explorations around
the food cup the still hungry rat was almost certain sooner or later
to stand up on its hind legs and rest its front paws on this bar,
which was so delicately pivoted that even a light downward pres-
sure would depress it. Moreover, the bar was so connected with
the food magazine that this downward movement would activate
the food-release mechanism with its characteristic sound vibration.
Because of the preceding training this vibration would at once cause
the animal to search in the cup for a pellet; however, no pellet
would be found, because in this phase of the experiment the food
magazine was empty. This being the case it might be supposed
that the act of pressing the bar would not be reinforced.

Skinner ran four rats through this experiment and is of the
opinion that the click really did reinforce the bar-pressing act as
contrasted with innumerable other acts evoked by the situation. A
record made by one of these animals is reproduced as Figure 18.
Each small unit in the rise of this curve represents one operation
of the lever by the rat. It may be seen that the reactions, as shown
by the slope of the curve, occurred with about maximum frequency
at the outset of the process and then gradually ceased as the curve

PRINCIPLES OF BEHAVIOR

became horizontal. Since no food was given, this learning is pre-
sumably the result of secondary reinforcement.

Unfortunately, Skinner publishes no account of an appropriate
control experiment to show how frequently a comparable animal
would depress the bar if the sound of the action of the magazine
had also been eliminated. Nevertheless, much corroborative evi-
dence from other investigations supports Skinner’s view that gen-
uine learning took place.
One such investigation is
reported by Bugelski ( 1 ).

Bugelski performed an
experiment in which he
trained two comparable
groups, each of 32 albino
rats, to press a bar for a
food-pellet reward in an
apparatus much like that of
Skinner. At the completion
of training, the bar-pressing
habits of both groups of
animals were extinguished 1 by so adjusting the apparatus that
pressure on the bar was no longer followed by the delivery of the
food pellet. With one group, however, the depression of the bar
was followed at once by the customary click of the food-release
mechanism, but with the other group it was not. Bugelski found
that the click-extinction group, as a whole, executed a little over
30 per cent more pressures on the bar before reaching extinction
than did the non-click group. This indicates in a convincing man-
ner the power of a stimulus (the magazine click) closely associated
with the receipt of food to contribute to the maintenance of a
receptor-effector connection at a superthreshold level.

The Skinner and Bugelski studies, taken jointly, enable us to
answer the first two of the questions raised by the Russian secon-
dary-reinforcement experiments. On the analogy of the Pavlovian
type of conditioned-reflex experiment, the vibrations of the food-
release apparatus in Skinner’s experiment may be assumed to have
become conditioned to salivary secretion and other phases of the

i The subject of extinction will be taken up in detail in a later chapter
(p. 258 ff.). For the present it may merely be said that when a learned
reaction is repeatedly evoked and the evocation is not followed by reinforce-
ment, the stimulus gradually loses its capacity to evoke the reaction. This
loss is known as experimental extinction.

Fia. 18. Record of secondary-reinforce-
ment learning in which the act learned
(pressure on a bar) was distinct from that
primarily conditioned (putting head down
to food cup, salivating, etc.). (After Skin-
ner, 9, p. 83.)

SECONDARY REINFORCEMENT 89

eating process, as, clearly, was the tendency to put the head down
to the cup and search for the food, by primary reinforcement. 1 On
the other hand, the act secondarily reinforced in the Skinner experi-
ment was that of depressing a bar. Two acts could hardly be more
diverse than sniffing in a food cup and depressing a bar with the
paws. We conclude, then, that the identity of the act in the pri-
mary and the secondary reinforcement of the Russian experiments
was an accidental condition and that such an identity is not a
necessary characteristic of secondary reinforcement in general.
Thus the first of the above questions is answered in the negative.

The two experiments just examined also enable us to answer
the second question posed by the Frolov experiment. Since the act
secondarily conditioned was clearly different from that conditioned
to the secondary reinforcing stimulus, it follows that the double
role of evoking the reaction to be conditioned and at the same time
serving as a reinforcing agent is not necessarily characteristic of
secondary reinforcing agents in general.

MUST THE SECONDARY REINFORCING STIMULUS EVOKE ITS
CONDITIONED REACTION IN ORDER TO ACT AS

A REINFORCING AGENT?

The finding of an answer to the third of our questions involves
a determination of the differential causal efficacy in secondary rein-
forcement of two events which usually occur together, namely, the
so-called secondary-reinforcing stimulus and its conditioned reac-
tion. The problem could be solved readily enough if we could
devise some way of presenting in a situation known to be capable
of reinforcement each of the factors separately, and observing
whether or not learning does in fact occur in each case. The diffi-
culty lies primarily in accomplishing the complete and certain
elimination of the one factor while retaining the complete integrity
of the other. While no investigations have been found which were
deliberately directed to a solution of this problem, there are two
or three which throw indirect light upon it. One of these, reported
by Cowles {2), purports to have eliminated from the secondary
reinforcement situation at least the gross overt reaction conditioned
to the secondary reinforcing stimulus.

Cowles readily trained two chimpanzees to insert 1%-inch col-
ored disks into a slot machine which delivered a raisin for each

1 However, see terminal note entitled, “The Status of Food Reward as a
Reinforcing Agent.”

90 PRINCIPLES OF BEHAVIOR

disk inserted. After this training the animals would retain, hoard,
and even expend considerable amounts of energy to secure the disks,
evidently as subgoals. Later each animal was presented with the
task of learning which of a row of five small, lidded boxes con-
tained concealed within it one of these tokens. Each learning
session consisted of 20 trials, all of which had to be completed
before the animal could exchange its tokens for raisins at the slot
machine which was located in a room about 35 feet distant. Each
series of 20 trials was devoted to a separate box, so that old habits
of choosing other boxes on previous training series had to be broken,
and new ones substituted, at every succeeding session. Finally on
alternate sessions, in order to secure a measure of the relative
strength of primary and secondary reinforcement, instead of a
token real food in the form of a raisin was put in the box. It was
found that the average score of the two apes on the second half
of each 20 trials on a given box (where chance success alone would
yield 20 per cent correct choices) was 74 per cent for food tokens
and 93 per cent for the food. This shows an amount of learning
due to secondary reinforcement which is fairly comparable to that

mediated by the primary food reward itself.

In this experiment the overt act normally evoked by the sec-
ondary reinforcing agent, the food token, was that of inserting the
token in the slot machine, whereas the act involved in the process
of secondary reinforcement was that of lifting the lid of a par-
ticular box in a row of five. Under the conditions of this multiple-
choice learning the slot machine was in a different room, and the
animals literally could not carry out the act of inserting the token
for some time after obtaining it. Moreover, Cowles reported no
tendency on the part of the apes to execute movements such as to
insert the disks into an imaginary vending machine. In so far
the Cowles experiment seems to yield a negative answer to the
third query suggested by Frolov's experiment; i.e., it suggests that
the occurrence of the first-order conditioned stimulus is necessary
for the setting up of a second-order conditioned reaction, and that
the occurrence of the primary conditioned reaction is not necessary.

the effect of EXTINCTION- of the primary receptor-

effector CONNECTION ON SECONDARY REINFORCEMENT

Despite the results from Cowles’ experiment, it would be rash
to conclude at once that the evocation of some fractional compo-

SECONDARY REINFORCEMENT 9 1

nent of the reaction originally conditioned to the secondary rein-
forcing stimulus was not present at each secondary reinforcement.
The principles of reinforcement learning lead a priori to the expecta-
tion that salivation, and probably many other hidden internal
processes such as the galvanic skin reaction, must have been con-
ditioned both to the stimulus energies arising from the vending
machine and to those from the food tokens. Moreover, we saw
above that in Frolov’s experiment salivary secretion accompanied
every secondary reinforcement of the black square. The fact that
such processes were not observed in Cowles’ experiment argues little
against the formidable probability that they did in fact occur. Had
not special apparatus and procedures been employed, the salivary
secretion would not have been observed in Frolov’s experiment
either. On the positive side, Cowles (2) reports that the apes em-
ployed in his experiments showed a marked tendency to put the
food tokens in their mouths. Wolfe (12, p. 16) reports that both
food and food tokens would elicit anticipatory lip-smacking activ-
ity, but a brass non-food token of the same shape would not. It
is clear from the above considerations that further evidence will
be required to determine whether the presence of the reaction com-
ponent of the usual secondary reinforcing situation is a necessary
condition for the occurrence of secondary reinforcement.

We saw above in connection with the Bugelski experiment (p.
88, footnote) that receptors which frequently evoke a conditioned
reaction without accompanying reinforcement presently lose the
power of evoking this reaction. Experimental extinction accord-
ingly offers a means of separating a secondary reinforcing stimulus
from the reaction to which it was conditioned while it was acquir-
ing its secondary reinforcing powers. This circumstance makes
possible the presentation of the former without the latter in close
temporal proximity to a reinforcible receptor-effector conjunction.
Such a combination of circumstances occurred in a quantitative
experiment reported by Grindley (3).

In one of his experiments Grindley placed young chickens at
the beginning of a four-foot runway. At the other end were placed
grains of boiled rice. Half of the chickens were permitted to find
their way down the runway and eat the rice. The other half were
likewise permitted to go down the runway, but found a plate of
glass placed a few inches above the rice which prevented them from
eating the grains. An index based on the time required by each
of the two groups of chickens to traverse the runway on each of

PRINCIPLES OF BEHAVIOR

the 12 successive trials is shown in Figure 19. Our interest is con-
fined mainly to the scores of the group of chickens which saw but
could not eat the rice. This curve shows that for the first four
or five trials the non-eating chickens gained in speed nearly as
fast as did the rewarded ones. During the subsequent trials, how-
ever, the non-rewarded chickens gradually lost their rate of loco-
motion until at the eleventh and twelfth trials they did scarcely

NUMBER or TRIALS

Fia. 19. Graphs showing in parallel the
course of the learning of young chickens
to traverse a straight four-foot runway by
primary and by secondary reinforcement
which was unaccompanied by the original
supporting primary reinforcement. (After
Grindley, S, p. 179.)

better than the chance per-
formance of the very first
trial.

These results of Grind-
ley’s experiment are dupli-
cated by Skinner’s record
(Figure 18), which is so
constructed that the flatten-
ing out of his curve to a
horizontal represents ex-
perimental extinction after
what appears to be a rather
abrupt learning. Pavlov
reports the same phenom-
enon in the conditioned-
reflex learning situation ;
extinction was encountered
by him especially when he
attempted to set up third
and fourth order condi-
tioned reflexes.

The experimental results
just considered, as well as
those from numerous con-
cordant experiments of both

conditioned-reflex and selective learning, all point to the same con-
clusion, namely, that a secondary reinforcing agent (in Grindley’s
experiment the visual stimulus presented by the rice grains) loses its
power of secondary reinforcement when it loses its power of evoking
the reaction conditioned to it at the time it acquired its power of
secondary reinforcement; moreover, it would appear to possess the
power of reinforcement only to the degree that it possesses the
power of evoking this reaction. It follows that in cases in which the
secondary reinforcing agent is still strong, secondary reinforcement

SECONDARY REINFORCEMENT

makes some progress. Very soon, however, it weakens through ex-
perimental extinction, reinforcement is thereby withdrawn, and the
secondary reinforcement declines along with the strength of its

parent receptor-effector connection.

Granting that the occurrence of at least some component of the
reaction conditioned to the stimulus in the usual secondary rein-
forcing situation is necessary for secondary reinforcement to occur,
our original question is still far from answered. Even if the reac-
tion is necessary, this does not mean that it is sufficient. This
confronts us with the question of whether the stimulus also is neces-
sary. The empirical solution of the latter problem would require
that the reaction conditioned to the stimulus of a secondary rein-
forcing situation somehow be presented without the evoking stim-
ulus, in close temporal proximity to a rcinforcible receptor-eflector
conjunction, and that a determination be made as to whether or
not learning takes place. No experiments in which this was at-
tempted have been found. Judgment in this intricate but theo-
retically important matter must accordingly be held in abeyance
until more adequate evidence becomes available.

the possibility of secondary reinforcements above the

SECOND AND THIRD ORDERS

A fourth question raised by the Russian experiments on secon-
dary reinforcement concerns the possibility of setting up condi
tioned reactions above the second order. In discussing this question
Pavlov remarks (7, pp. 34-35) :

It was found impossible in the case of alimentary reflexes to press
the secondary stimulus into our service to help us in the establishmen
of a new conditioned stimulus of the third order. Conditioned reflexes of
the third order can however be obtained with the help of t e secon
order of conditioned reflexes in defence reactions such as that against,
stimulation of the skin by a strong electric current. But even m this
case we cannot proceed further than a conditioned reflex of the third order.
• • . In these conditioned reflexes, passing from the first to the third order,
the latent period progressively increases. In the same or er we pass
from the strongest to the weakest conditioned defence reflex.

The above passage from Pavlov has been quoted at length be-
cause there is reason to believe that while the facts there reported
are well authenticated, their adaptive implications have occasion-

PRINCIPLES OF BEHAVIOR

ally been misunderstood, possibly even by Pavlov himself. This
presumptive error in interpretation seems to have come about
through the distinctly narrow and artificial nature of the condi-
tioned-reflex experimental procedure by means of which it was
investigated, much as is believed to have been the case in the deter-
mination of what constitutes a primary reinforcing state of affairs
(p. 78). In order to prove that secondary reinforcement is a
genuine phenomenon it is necessary to remove all possibility of
primary reinforcement from the situation. This, of course, inciden-
tally causes extinction of the primary receptor-effector connection,
which, as we saw in Grindley’s experiment, soon leads to the loss
of the power of reinforcement by the secondary reinforcing agent
and thus tends to bring the potential chain of transfers of the power
of reinforcement to an early termination. As indications of this
we note Pavlov’s statement that each successive reaction in the
chain had a progressively longer latency, itself an indication of
receptor-effector weakness which is also separately noted by him.
Viewed in the light of these considerations the limitation in the
number of higher-order conditionings was presumably a mere arti-
fact of the technical procedure employed in the investigation of

the problem.

We now have evidence which indicates that the gradient of
reinforcement does not extend backward from the reinforcing state
of affairs in detectable amounts beyond about 20 seconds (5) . This
means that in protracted behavior sequences, even with primary
reinforcement fully intact, the direct effects of the latter are auto-
matically excluded for all receptor-effector conjunctions beyond a
half minute or so from the point of such reinforcement. Since in
the higher organisms behavior sequences, e.g., the pursuit of prey,
often continue far beyond such a limit, it follows that the strength
of the earlier segments of such sequences must frequently be main-
tained by very long chains of secondary reinforcing situations. In
normal human organisms it would appear from such considerations
that higher-order conditioning yields receptor-effector connections
which are comparable in vigor with those produced by primary
reinforcement, and that there is practically no limit to the degree
to which higher-order conditioning may be carried under suitable
conditions. As for the latter, it may be said that with the excep-
tion of the nature of the reinforcing state of affairs involved the
conditions necessary for secondary reinforcement are the same as
those for primary reinforcement. This is to say that a receptor

SECONDARY REINFORCEMENT 95

impulse will acquire the power of acting as a reinforcing agent if
it occurs consistently and repeatedly within 20 seconds or so of a
functionally potent reinforcing state of affairs , regardless of whether
the latter is primary or secondary.

THE ROLE OF SECONDARY REINFORCEMENT IN COMPOUND

SELECTIVE LEARNING

As a means of showing something of the systematic and func-
tional significance of secondary reinforcement we shall now con-
sider a few of its more elementary implications. It has already
been suggested (p. 84) that secondary reinforcement plays its
major role in protracted behavior sequences, particularly in those
portions of them which precede the point of primary reinforcement
by more than 20 seconds or so. A typical sequence of this kind is
found in compound selective learning. This may be thought of as
a series of trial-and-error learning situations in which each link
in the series consists of a simple selective learning situation pos-
sessing the general characteristics of Demonstration Experiment A.
It may further be assumed that at the beginning of learning 20
seconds or more are consumed by the organism before the correc
reaction is performed at each of five choice points; that the correct
reaction in one situation always leads at once to the next situation
in an invariable order; that the new situation instantly activates
the receptors of the organism in a distinctive way ; and that the
entire sequence finally culminates in the complete satiation of the
need which motivated the organism throughout the total activity,
following common-sense usage, we shall call this final primary
reinforcing state of affairs the goal and represent it by the letter G.
If there are five segments in such a behavior sequence, we may
represent the correct reaction of the first trial-and-error situation
by R t) that of the second by R t , and so on to R 5 . In a similar
manner the gross stimuli presented to the organism by the respec-
tive situations may be represented by the parallel notation, S ly £>t,

S», and S 6 . - . > •

It follows from the conditions of compound selective learning

assumed above that the organism by sheer trial-and-error wi

work blindly down through the series until S 5 — > evokes R s ,

which by physical causation produces G, the primary reinforcing

•state of affairs. This, by the principle of primary reinforcement,

begins to set up the connection S s — — » s 5 > Rs and at the same

9 6 PRINCIPLES OF BEHAVIOR

time, by the principle of secondary reinforcement, to endow S$ with
the powers of (indirect) reinforcement. Now, R+ is too remote
in time from G to be selected by primary reinforcement. However,
in the course of subsequent executions of the sequence by the organ-
ism the secondary reinforcing power recently acquired by S 5 will
mediate the connection S^ » s^ — —^R^, at the same time endow-
ing S k with secondary reinforcing power. In the same way sec-
ondary reinforcement will move progressively backward from S u to
S 3f from S s to S t , and finally from S* to S If ultimately resulting in
the tightly knit, errorless behavior sequence shown in Figure 20.

It is clear from the preceding that compound selective learning
must necessarily progress in a backward manner from the point
of primary reinforcement. This means that in the situation under
consideration errors would be eliminated most quickly at S 5 , next

s,— v— v v r 2~ s 3 — *r~ s 4~ v— R 4 — s 5 —

Fio. 20. Diagrammatic representation of the final phase of a case of
compound selective learning in which, simple trial-and-error occurs five times.
If at the beginning of learning the occurrence of each correct R be assumed
to consume at least 20 seconds then the four segments at the left of the
figure would be dependent for their selection purely upon secondary rein-
forcement. In that event Si would be the secondary reinforcing agent effect-
ing the selection of R if and St would be the secondary reinforcing a g eD *'
effecting the selection of Ri. As in Figures 13, 14, and 16, the sign *

represents a non-physiological causal connection.

most quickly at S+, and most slowly of all at S ti the rate-of-leam-
ing score at the several points presenting a gradient whose highest
point would be at S s . That such a gradient exists in compound
selective learning situations is well known, though it may be over-
ridden by various known factors. Because of the relation of this
gradient to the point of primary reinforcement or goal, the back-
ward order of the elimination of errors has been called the goal
gradient (5). We shall recur to this subject in another connection
(p. 142).

There should also be noted the implication in the above analysis
that certain circumstances in the learning situation may effectively
prevent learning from occurring, particularly in the segments an-
terior to S s > s 5 > R 5 . One of the most important of these is

a delay in the occurrence of S s , say, following reaction R t . The
principles developed earlier in the chapter imply that in an ideal
situation in which S If S t , S 3f S if and S 5 are all totally differentia
delay of 20 seconds or more will prevent learning at any point m

SECONDARY REINFORCEMENT

the series anterior to On the other hand, if S lf S t) S Sf and
all contain an important component found in S 5 (a situation easily
set up experimentally) then each will become a reinforcing agent
regardless of the temporal gaps in the series.

SUMMARY

Primary reinforcement, because the range of its gradient is
limited to 20 or 30 seconds, is incapable of explaining the learn-
ing which manifestly occurs when the receptor-effector processes
involved are temporally very remote from the relevant need reduc-
tion. These latter reinforcements are explained by the discovery
that the power of reinforcement may be transmitted to any stimulus
situation by the consistent and repeated association of such stim-
ulus situation with the primary reinforcement which is character-
istic of need reduction. Moreover, after the reinforcement power
has been transmitted to one hitherto neutral stimulus, it may be
transferred from this to another neutral stimulus, and so on in a
chain or series whose length is limited only by the conditions which
bring about the consistent and repeated associations in question.
The inability of Pavlov and his pupils to obtain conditioned reac-
tions above the third order appears to have been due to the highly
artificial nature of their experimental procedures which did not pro-
vide the necessary conditions for long and stable secondary-rein-
forcement chains.

Our detailed findings concerning secondary reinforcement may
be listed as follows:

1. Perhaps the most striking characteristic of secondary reinforce-
ment is that it is itself a kind of by-product of the setting up of a
receptor-effector connection, in the first instance through primary rein-
forcement. Primary reinforcement, on the other hand, appears to be a
native, unlearned capacity in some way associated with need reduction.

2. Secondary reinforcement may be acquired by a stimulus from
association with some previously established secondary reinforcement, as
well as with a primary reinforcement. It would appear that transfer of
this power of reinforcement from one stimulus situation to another may
go on indefinitely, given the conditions of stable and consistent association.

3. A receptor-effector conjunction involving any effector may be rein-
forced by any secondary reinforcing situation. . ,

4. Secondary reinforcement differs from primary reinforcement in tha„
the former seems to be associated, at least in a molar sense, with stimula-
tion, whereas the latter seems to be associated with the cessation ot

stimulation, i.e., of the S D . # .

5. Stimuli which acquire secondary reinforcing power seem always to

98 PRINCIPLES OF BEHAVIOR

acquire at the same time a conditioned tendency to evoke an associated
reaction. The available evidence indicates that as such stimuli lose
through extinction the power of evoking this reaction, they lose m about
the same proportion their power of secondary reinforcement. It is
probable also that the reverse is true; that a stimulus gradually acquires
its powers of secondary reinforcement as it acquires the power of evoking

the reaction conditioned to it. . . A .

6 It follows from ( 5 ) that a stimulus alone is ineffective as a secondary

reinforcing agent. We do not know whether the reaction evoked by this
stimulus, or possibly some critical fraction of that reaction if presented
without the stimulus, would serve as a reinforcing agent. Conceivably, ot
course the combination of both the stimulus and its conditioned reaction,
or some fractional component thereof, is necessary to effect secondary

reinforcement. . ... ,

7 . It is apparent from the preceding that a reaction conditioned to a

stimulus which has as a by-product acquired secondary reinforcing power,
at a point several secondary-reinforcement links removed from the point
of primary reinforcement, may suffer experimental extinction in two ways:
(a) through the evocation of the reaction not being followed by the parent
secondary-reinforcing stimulus, and (6) through the evocation of the
reaction being followed by the parent stimulus after the latter has lost its
secondary reinforcing power through, say, the extinction of its own con-
ditioned reaction.

8. With the main facts of secondary reinforcement before us we may
now reformulate the law of reinforcement in such a way as to include the
wider learning potentialities inherent in secondary reinforcement:

Whenever an effector activity occurs in temporal contiguity
with the afferent impulse , or the perseverative trace of such an
impulse, resulting from the impact of a stimulus energy upon a
receptor, and this conjunction is closely associated in time with
the diminution in the receptor discharge characteristic of a need
or with a stimulus situation which has been closely and consistently
associated with such a need diminution, there will result an incre~
ment to the tendency for that stimulus to evoke that reaction.

NOTES

The Status of Food Reward as a Reinforcing Agent

We have seen in the preceding pages that primary reinforcement or ^ n ^'^
in the reduction of a primary need. Now, the primary need in the case of food
privation presumably consists in the requirement of the cells of the body o
nutriment, just as the primary need in the case of the shocked animate m Demon-
stration Experiments A, B, and C was the cessation of the action of the f
current on the cells of the animals* feet. It is evident from a knowledge of nuto
tional physiology that there is an appreciable delay between the w

mastication of food and the ultimate reduction m the nutrient need of t

SECONDARY REINFORCEMENT

cells while mastication, deglutition, digestion, and absorption are taking place.
This makes it distinctly improbable that the presentation of food, its mastication,
and the gustatory stimulus incidental to mastication are primary reinforcing
states of affairs. Yet innumerable experiments have shown that the presentation
and mastication of food in fact constitute a powerful reinforcing combination.
These considerations strongly suggest that the eating of food as such brings about
learning through secondary reinforcement rather than through primary reinforcement.

Incidentally, this hypothesis explains the paradox that whereas other cases of
clear primary reinforcement appear to be associated with reduction in drive
stimulation ( So —* sd ), food reinforcement appears at the instant of reinforcement
to be associated rather with stimulation or the onset of stimulation which, as we
have seen above, is characteristic of secondary reinforcement. This may account
for Pavlov's view that reinforcement is primarily a phenomenon of stimulation.

A Possible Technique for Determining the Status of Food Reward as a

Reinforcing Agent

Because of the wide use of food reward as a reinforcing agent in behavior
experiments its status in this respect is in especially urgent need of clarification.
There is reason to believe that a sham feeding experiment might contribute
materially to this end. The esophagus of a dog could be severed and both the
upper and lower sections converted into fistulas opening through the skin of the
dog’s neck. With such an arrangement it would be possible to feed the dog and
either have the masticated food fall into a conveniently placed receptacle or have
it pass through a tube connecting one fistula with the other and thus enter the
Btomach and ultimately reduce the need of the body cells for nutriment Since
the various receptor discharges associated with the eating of food, its swallowing,
digestion, and absorption, have throughout the entire life of each organism been
associated in a uniform and practically invariable sequence with ultimate need
reduction, it is to be expected that the stimuli associated with mastication would
have acquired a profound degree of secondary reinforcing power For this reason
the power of secondary reinforcement ought not to be lost by such stimuli throug
a moderate amount of experimental extinction. However, if the present hypothe-
sis is sound, a dog sham fed on one kind of food and really fed on an equally rein-
forcing kind of food should, after a time, show a distinct preference for the food
which mediates nutrition and so primary reinforcement; after much training it
might even refuse to eat the sham food since, not being reinforced this activity
should suffer experimental extinction. Careful controls would, of course need
to be carried out on such matters as the food preferences possessed by the dog
just before the beginning of the experiment. Numerous variants of the above
procedure will at once suggest themselves as alternatives in the solution of this

extremely important problem.

Are Primary and Secondary Reinforcement at Bottom Two Things or One?

'•> In the present chapter it has been seen that despite well-marked differences
there are a number of striking similarities between primary and secondaiy re-
inforcement. Perhaps the most notable of the similarities is the act o rein orc^
ment itself. So far as our present knowledge goes, the habit structures mediated
by the two types of reinforcement agents are qualitatively 1 entica . s
consideration alone constitutes a very considerable presumption in favor of the

IOO

PRINCIPLES OF BEHAVIOR

view that both forms are at bottom, i.e., physiologically, the same. It is difficult
to believe that the processes of organic evolution would generate two entirely
distinct physiological mechanisms which would yield qualitatively exactly the
same product, even though real duplications of other physiological functions are
known to have evolved.

While the ultimate proof of the essential identity of the two processes, when
and if it comes, must be looked for on the physiological rather than on the be-
havioral level, it is evident that the present development of neurophysiology
is quite remote from such an achievement. Meanwhile the urgency of the prob-
lem from the standpoint of systematic behavior theory is such as to make an
attempt at a workable first approximation to such a proof on a molar level ex-
tremely desirable. This would necessarily take the form of a revised inductive
generalization which would be sufficiently comprehensive to include both phe-
nomena as special cases of the same rule or law. While no detailed and fully
substantiated hypothesis of this nature will be attempted here, a few suggestions
are offered which may possibly contribute to the attainment of this end by others.

In the development of this plan the first important fact to be considered is that
the initial secondary reinforcing stimulus acquires its reinforcing power through a
process of reinforcement. Moreover, the process of successive transmission of
this power from one stimulus situation to another, backward through a compound
selective learning sequence, also appears always to occur in a reinforcement
situation in which the secondary reinforcing stimulus acquires a reaction tend-
ency. These considerations suggest rather strongly that the first secondary
reinforcing stimulus acquires its power of reinforcement by virtue of having
conditioned to it some fractional component of the need reduction process of the
goal situation ( G , Figure 20) whose occurrence , wherever it takes place , has a specific
power of reinforcement in a degree proportionate to the intensity of that occurrence.

Let us represent thisjfractional component of the goal reaction by the symbol g.
Referring back to the compound selective learning situation represented in
Figure 20, it is evident on the above hypothesis that while Si — ► 8t is acquiring
its connection to Rt, the peraeverative trace of a* is also acquiring a parallel connec-
tion to g. It follows that presently, when the connection Si — * a* ► g acquires a

superthreshold strength, g, by the principle of stimulus generalization (see p.
183 ff.), will come forward in the series and occur in close conjunction with St

and will therefore serve to strengthen S* — ► a* ► R t as well as to condition itself

to Si — i ► a«, thus: — * »4 ► g. In the course of successive trials or repetitions,

when this connection (a 4 ► g) becomes of superthreshold strength, the same thing

would occur with S t , then with S 3 , and finally with S%. In this way g, as both a
conditionable process and a reinforcing agent, would be passed back through
the sequence. There would therefore develop an understandable modus operand i
for compound selective learning.

Since it takes time for g to become conditioned to a new trace to a degree such
that it can act with much strength as a reinforcing agent, there is bound to be
considerable delay between its full action at the goal and that at the beginning
of the behavior sequence leading to the need reduction. This would, of course,
produce the goal gradient, now known to be characteristic of such learned se-
quences ( 10 , 11).

We now turn to the matter of experimental extinction. Suppose, m case
selective learning has occurred at all the five choice points, that the organism
executes the complete action sequence, but that the primary reinforcing state of

SECONDARY REINFORCEMENT

ioi

affairs, G, does not follow the performance of Rs. It must be supposed that the
conditioned g as evoked by Ss — * ss will be weaker than the unconditioned g
evoked in connection with the need reduction at G. Therefore it is to be expected

that while Ss —* ss » g will somewhat retard the process of extinction of — »

Ss ► Rs (as in Bugelski’s experiment, 1), it will nevertheless not suffice to prevent

ultimate extinction. In a similar manner the weakening of Ss — ► s 5 * Rs will

rapidly bring about the weakening of S« — ► s 4 * g; this will result in a weakening

of S 4 — * s« ► Ri, and so on backward throughout the sequence to its very begin-
ning, gradual collapse of the entire behavior sequence occurring as trial follows
trial during the extinction process. Presumably generalized extinction effects
would contribute to the speed of inhibition throughout such a behavior series,
especially where the trials are given in immediate succession (4, pp. 497-499).

It is to be noted in this connection, however, that the present hypothesis
does not imply that secondary reinforcement will necessarily suffer experimental
extinction when the support of the primary need reduction is withdrawn. If the
primary reinforcement has been sufficiently profound for the connection Ss — *

Ss > g to be very strong, the conditioned g may be intense enough to withstand

the inhibition generated by an indefinitely large number of otherwise unreinforced
presentations of S t , S it or Ss. Here, apparently, we have the explanation of what
Gordon Allport has called the functional autonomy of higher-order conditioned
reactions; the g, if sufficiently well conditioned, may be strong enough in rein-
forcing power to maintain itself through self-reinforcement — a true functional
autonomy. It is probable that something of this kind is operative in certain
cases of neurotic symptoms, as has been pointed out by Mowrer ( 6 ) in his be-
havioristic interpretation of Freud’s doctrine of anxiety.

REFERENCES

1. Bugelski, R. Extinction with and without, sub-goal reinforcement J

Comp. Psychol. , 1938, 26, 121-133.

2. Cowles, J. T. Food-tokens as incentives for learning by chimpanzees.

Comp. Psychol. Monog., 1937, 14, No. 5.

3. Grindley, G. C. Experiments on the influence of the amount of reward

on learning in young chickens. Brit. J. Psychol., 1929, 20, 173-180.

4. Huix, C. L. Goal attraction and directing ideas conceived as habit

phenomena. Psychol. Rev., 1931, 38, 487-506.

5. Hull, C. L. The goal gradient hypothesis and maze learning. Psychol.

Rev.) 1932, 39 f 25—43.

6 . Mowrer, O. H. A stimulus response analysis of anxiety and its role as a

reinforcing agent. Psychol. Rev., 1939, 46, 553-565

7. Pavlov I P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Umv. Press, 1927.

8 * investigation of the delay-of-reinforcement

gradient. PhD. thesis, 1942. On file Yale Univ. Library.

Centu% Co InI l9 3 f W ° r Y ° rk: Apple '°“-

10 ' bliQds in maze learning by the

re "' ard up ° n IearniDS in the white rat

12 - 5° kC “ dS f0r Ch,mPan2CM -

CHAPTER VIII

The Symbolic Construct S H R as a Function of the

Number of Reinforcements

In the course of the preceding discussions of reinforcement the
reader may have noticed two implicit assumptions: (1) the recep-
tor-effector connections so set up correspond roughly to what are
known to common sense as habits; 1 and (2) the process of habit
formation consists of the physiological summation of a series of
discrete increments, each increment resulting from a distinct recep-
tor-effector conjunction (sC R ) closely associated with a reinforcing
state of affairs (G). It shall be our task in the present chapter to
try to tease out of a series of relatively simple and reasonably
well-studied habit-formation situations at least a first approxima-
tion to the central law or functional relationship of habit strength
as dependent upon the number of these reinforcement increments.

As a preliminary to this undertaking it is important to note
that habit strength cannot be determined by direct observation,
since it exists as an organization as yet largely unknown, hidden
within the complex structure of the nervous system. This means
that the strength of a receptor-effector connection can be deter-
mined, i.e., can be observed and measured, only indirectly. There
are two groups of such observable phenomena associated with
habit: (1) the antecedent conditions which lead to habit formation,
and (2) the behavior which is the after-effect or consequence of
these antecedent conditions persisting within the body of the organ-
ism. As our analysis progresses we shall find that habit strength
depends upon various antecedent factors in addition to the number
of reinforcements. We shall also note that habit strength may
manifest itself in several different measurable ways. One of these,
the magnitude of the evoked reaction, will next be considered.

i Strictly speaking, by common usage the referent of the term “habit
is a well-worn mode of action, whereas by the present usage the referent xs
persisting stale of the organism (resulting from the remforcement) wh^ch
a necessary, but not a sufficient, condition for the evocation of the act

in question.

102

sHh and the number of reinforcements

io 3

HABIT STRENGTH AND REACTION MAGNITUDE

The progressive increase in the magnitude of an evoked habitual
reaction with successive reinforcements is conveniently illustrated
by a study reported by Hovland ( 1 ). This investigator associated
the Tarchanoff galvanic skin reaction, originally evoked by a mild
electric shock on the wrist, with the simple sinusoidal vibrations of

0 8 16 24 32 40 48

. HUMBER Of REINFORCEMENT REPETITIONS (N)

Fia. 21. An empirical learning curve plotted in terms of the amplitude in
millimeters of the first galvanic skin reaction evoked by the conditioned
stimulus after varying numbers of reinforcements. The circles represent mean
readings for different but comparable groups of subjects. The curved line
was plotted from values secured by substituting in an equation fitted to the
data represented by the circles. (From data published by Hovland, l, p. 268.)

a beat-frequency oscillator. The galvanic reaction was picked up
from the skin of the subject’s hand by a pair of polished silver disk
electrodes, one bound to the palm and the other to the back. The
current thus secured was passed through a sensitive galvanometer,
the magnitude of the subject’s electrical reaction being shown by
the amount of movement across a screen made by a beam of light
reflected from a mirror in the instrument. This movement was
recorded graphically and was subsequently measured in millimeters.
Four matched groups of 32 subjects each were given 8, 16, 24, and
48 reinforcements respectively. At once following this series of

PRINCIPLES OF BEHAVIOR

104

reinforcements the tone was presented alone, and the amplitude (A)
of the evoked reaction was measured. The means of the condi-
tioned reactions thus evoked from the several groups of subjects
are shown by the circles in Figure 21. The circle at zero is the,
average amplitude of reactions evoked from all groups of subjects
previous to reinforcement. The curved line drawn among the data
points represents a series of values calculated from a growth func-
tion which has been fitted to the data represented by the circles.

From an inspection of Figure 21 three observations may be
made:

1. The value of this reaction at zero reinforcements is not itself zero
but has, on the contrary, the very appreciable value of 3.16 millimeters.
This is quite characteristic, since it has long been known that the
galvanic skin reaction is evocable in considerable amounts by any stimulus
of appreciable intensity.

2. The greater the number of reinforcements (and, presumably, the
stronger the habit), the greater will be the amplitude of the evoked
reaction. Accordingly, the amplitude of the reaction is said to be an
increasing function of the number of reinforcements. 1

3. Despite a certain amount of deviation of the circles from the fitted
curve, possibly due to the limited number of data from which the means
were calculated, the relationship appears to approximate rather closely a
simple positive growth function.

HABIT STRENGTH AND REACTION LATENCY

A second way in which habit strength may manifest itself in a
measurable manner is in the length of time elapsing from the onset
of the stimulus to the onset of the associated reaction (at#). This
time interval is called reaction latency.

The general relationship of habit strength (as indicated by the
number of reinforcements) to reaction latency is illustrated by an
investigation reported by Simley (5). In this study college stu-
dents associated nonsense characters, presented for five seconds
each by means of an automatic exposure apparatus, with nonsense
syllables presented orally by the experimenter in the middle of
each exposure. The subjects were instructed to speak each syllable
just as quickly as possible after the corresponding character was
presented. A voice key connected with other automatic devices
made possible the determination of each reaction latency as the

1 This statement holds for certain reactions, such as salivary secretion
and the galvanic skin reaction, but apparently not for all (see p. 329£T.L

aHa and the NUMBER OF REINFORCEMENTS 105

learning progressed. Learning was continued long after the habits
involved in any given rote series had passed the reaction threshold.
Out of a large number of such stimulus-response combinations, one
subject (M.W.) was found
to have spoken the asso-
ciated syllable before being
prompted at the second pre-
sentation of about 125 of
the characters. The mean
latencies of these reactions
at the second and each of
the following fifteen rein-
forcements 1 2 * are shown by
the circles in Figure 22. As
in the case of Figure 21, a
function has been fitted to
these data; this is repre-
sented by the curve which
passes among the circles.

An inspection of Figure
22 shows:

1. There is no circle rep-
resenting a latency value at
zero reinforcement; this is be-
cause no reaction of this com-
plex type can occur previous
to any learning. Strictly speak-
ing, this means that the la-
tency of such a reaction is
infinite when reinforcement is
zero.

2. Reaction latency is in-

versely related to the number

* 4 v # 10 c % »

SUCCESSIVE REINFORCEMENTS (N)

Fia. 22. Empirical learning curve plotted
in terms of the reaction latency of speaking
nonsense syllables at the presentation of the
nonsense character with which they were
paired. The circles represent means from
about 125 such syllable reactions by a single
subject (M.W.) at 16 successive character
presentations and after as many reinforce-
ments. The curved line represents the re-
ciprocal of a fractional power of a positive
growth function fitted to the data (see
equation 5, p. 121). (From results pub-
lished by Simley, 6.)

of reinforcements; i.e., the .

greater the number of reinforcements (and, presumably, e s g
habit), the shorter the time required for reaction evocation. Thus r ®^ tl0n
latency is said to be a decreasing function of the number of remforce-

mei 3. S Quite as in Figure 21, the data values deviate appreciably from the

1 Reinforcement in such learning is evidently secondary. (See Chapter

This statement holds for certain conditions of training but not for all
(see p. 337).

io 6

PRINCIPLES OF BEHAVIOR

fitted curve, though in general the deviations are not excessive. The
relationship thus appears to approximate rather closely the reciprocal of a
positive growth function.

HABIT STRENGTH AND RESISTANCE TO EXPERIMENTAL

EXTINCTION

A third way in which habit strength may manifest itself in a
measurable manner is in its resistance to the effect of repeated
evocations unaccompanied by reinforcement, which ordinarily pro-
duces experimental extinction (see Chapter XV). An illustration
of this functional relationship is found in a study by Williams ( 6 ),

bar-pressing reactions evoked by the conditioned stimulus in different groups
of hungry albino rats after varying numbers of food reinforcements. The
curved line represents a positive growth function which has been fitted to the
data represented by the circles. (Data from Williams, 6 , and Penn; figure
adapted from one published by Perin, 4 .)

as supplemented by Perin (4). These investigators trained groups
of hungry albino rats to depress a bar in order to secure food pellets
in an apparatus resembling that of Skinner and Bugelski, described
above (pp. 87, 88 ff.). The several groups of animals were given
varying numbers of reinforcements, after which the reactions were

a Hs AND THE NUMBER OF REINFORCEMENTS 107

no longer followed by the food reward. The mean number of
unreinforced reactions (n) which were made by the respective
groups of animals before an interval of five minutes or more
occurred between two successive responses was taken as the measure
of resistance to extinction . 1 The relationship of these values to
the number of reinforcements is shown by the circles in Figure 23.
The curved line running through these data points represents a
positive growth function fitted to them.

An inspection of this figure shows:

1. At zero reinforcements the number of unreinforced reactions re-
quired to produce extinction is negative. This is a quantitative expression
of the fact (noted above in connection with reaction latency, p. 105)
that a habit must have a certain strength before any reaction at all can
be evoked (see Chapter XVIII), and therefore before any directly measur-
able extinction effects can possibly be observed. This negative value ( — 4)
was obtained indirectly by extrapolating backward to zero reinforcements
the function fitted to the values obtained from superthreshold strengths
of the habit.

2. The greater the number of reinforcements (and, presumably, the
stronger the habit), the greater will be the number of non-reinfo reed
reactions required to produce a given degree of experimental extinction.
Resistance to experimental extinction may therefore be said to be an
increasing function of the number of reinforcements.

3. In general the data points, in spite of the usual deviations from
the fitted curve, approximate rather closely a simple positive growth
function, as we saw to be the case with reaction amplitude.

HABIT STRENGTH AND PER CENT OF CORRECT REACTION

EVOCATION

A fourth way in which an increase in habit strength may mani-
fest itself is by a change in the per cent of occurrences of the
stimulating situation which evokes the reinforced reaction. This
functional relationship is illustrated by the results of an unpub-
lished experiment performed by Bertha Iutzi Hull. An albino rat
was presented with a pivoted rod projecting through a cross-shaped
aperture in the side of a restraining box. The hollow end of the
rod was filled with sticky food. In securing this food the animal
incidentally moved the rod more or less at random into all four
arms of the cross, but into some much more frequently than others.
At first the apparatus was set to give the rat a small pellet of

1 An alternative and closely related measure of this same function is the
time required to produce experimental extinction (.A).

io8

PRINCIPLES OF BEHAVIOR

food whenever the rod was moved in any one of the four directions.
After this had determined the relative strength of the animal’s

reaction tendencies into each
of the four arms of the cross,
the end of the rod was care-
fully cleaned and the appa-
ratus set to give food pellets
only when the rod was moved
in the hitherto least preferred
direction. Owing to the large
number of trials required for
this bit of learning, the ele-
ment of chance was largely
eliminated from the results
and a very smooth curve of
Fig. 24. Empirical learning curve the process was obtained
plotted in terms of the per cent of cor- from single anima l. This
rect (reinforced) reactions by an albino . .

rat from a group of four possible direc- 1S shown in Figure 24. ine

tions of movement of a rod projecting per cent of correct reactions
through a cross-shaped aperture (Plotted for successive groups 0 f 100
from an unpublished study by Bertha ... .

Iutzi Hull.) trials is indicated by tne

circles. Since no equation
has been fitted to these data, the circles have merely been con-
nected with straight lines.

An examination of Figure 24 shows:

1. The greater the number of trials (and, presumably, the greater
the relative habit strength of the reinforced reaction), the greater will
be the per cent of correct reactions.

2. The curve begins with a relatively brief period of positive accelera-
tion.

3. The period of positive acceleration is succeeded by a relatively pro-
tracted period of negative acceleration. This had not reached the maxi-
mum of 100 per cent correct reactions when the experiment was
terminated.

4. The combination of the positively and the negatively accelerated
portions of the learning curve gives it a definitely sigmoid shape which
is strikingly different from that of the three learning curves previously
considered.

THE CONCEPT OF HABIT STRENGTH AS SUCH

We have seen exemplified four cases of relatively simple habit
formation. In all these cases it has been assumed that habit

sHs AND THE NUMBER OF REINFORCEMENTS 109

strength has progressively increased with the number of reinforce-
ments. On this assumption, the progressive increase of habit
strength in each of the four learning situations is manifested as
a distinct measurable function of the number of reinforcements.
While no habits manifest themselves in all of the four manners at
every point of the acquisition process, most do so at one stage
or another. No habits mediate overt action below the reaction
threshold (p. 323 ff.). On the other hand, all well-established habits
display on the presentation of the relevant stimulus a certain mag-
nitude of reaction, a certain reaction latency, and a certain resist-
ance to experimental extinction. Moreover, many habits at or near
the reaction threshold (see Chapter XVIII), as well as all processes
of selective learning at medium strengths, display a progressive
increase in the frequency with which the stimulus evokes the reac-
tion being reinforced. Indeed, certain habits, e.g., the conditioned
lid reaction, may manifest themselves in all four ways at some
particular stage of their formation.

From the above considerations, as well as from everyday obser-
vation, it is clear that at any time for months and years following
specifiable reinforcement the presentation of the conditioned stim-
ulus is likely to evoke the reaction. Now, it is assumed that the
immediate causes of an event must be active at the time the event
begins to occur. But at the time a habit action is evoked, the
reinforcing event may be long past, i.e., it may no longer exist;
and something which does not exist can scarcely be the cause of
anything. Therefore, reinforcement can hardly be the direct or
immediate cause of an act. We accordingly conclude that the
immediate cause of habit-mediated action evocation must be a com-
bination of (1) the stimulus event and (2) a relatively permanent
condition or organization left by the reinforcement within the
nervous system of the animal. This last is what is meant by the
term habit. It will evidently be a decided convenience to speak
of this persisting physical condition as distinct from either the
reinforcing events which produced it or the overt activities which,
on occasion, it may itself mediate.

In this connection it may be recalled from the preceding pages
that the after-effects of reinforcement manifest themselves in mul-
tiple modes. From a positivistic point of view, the question natu-
rally arises as to which of these, if any, is to be considered the
index of habit strength. The fact seems to be that no one of them

I IO

PRINCIPLES OF BEHAVIOR

merits this distinction more than the others. A careful survey of
the evidence has led to the belief that while habit strength is the
dominant factor determining the amount of each of the four aspects
of reaction due to reinforcement after-effects, the latter are acting
in conjunction with some other and distinct factors in each case.

What those factors may be, experimentalists are at present
busily engaged in finding out, but some things are already known.
It is a matter of common observation that we learn by trial and
error specifically to react violently, moderately, or gently as a
given situation requires in order that reinforcement shall occur
(p. 304 ff.). Thus magnitude of reaction may he learned as such.
Also, Pavlov (5) long ago showed that reactions could be condi-
tioned to various delays by suitable methods of training. Thus
latency also may be learned as such. Finally, whether or not a
reaction will be evoked by a given stimulus apparently depends not
only upon the strength of the original reinforcement but upon the
other stimuli which may be acting at about the same time; upon
the strength of any competing reaction tendencies associated with
the other stimuli; and upon a very large number of additional fac-
tors, some of which will be elaborated in subsequent chapters (p-
341 ff.). In the calculation of the magnitudes of these various
manifestations as joint functions of habit strength it will clearly
be a convenience to have a single value to represent the influence
of the several factors which act together to determine its amount.

It is quite possible, of course, that a theory of behavior could
be developed without employing the concept of habit strength. In-
deed, it is probable that all of the many constructs (such as elec-
trons, protons, etc.) which are employed to represent unobservables
in the physical sciences could be dispensed with if scientists cared
to resort to the use of expressions sufficiently complex to represent
explicitly all the observations from which the nature and amounts
of the unobservable entities are inferred. When the relevant ob-
servations upon which the presence and magnitude of the unobserv-
able entity depend are properly represented by a single number
or sign, they can all be manipulated at once quite adequately by
the manipulation of that symbol. Indeed, this is a routine practice
in mathematics, where such conditions obtain. To repeat every
one of them on each occasion in which the group of factors as a
whole is to be manipulated would be a pedantic waste of effort.
On the other hand, if the system has been properly constructed the

sHm and the number of reinforcements

1 1 1

sign can at any time be expanded by an explicit representation of
the various factors for which it stands. This manoeuvre, of course,
converts the system substantially into positivistic form; which
shows that fundamentally the use of constructs, where permissible
(see terminal note), is no different than an ordinary positivistic
procedure such as that advocated by Woodrow (7).

The use of logical constructs thus probably in all cases comes
down to a matter of convenience in thinking, i.e., an economy in
the manipulation of symbols. It is accordingly on the ground of
convenience and economy rather than of strict necessity that the
attempt is here being made to retain the substance of the common-
sense notion of habit strength. Students of behavior who have a
positivistic distaste for logical constructs may adapt the present
systematic approach to their own preferred mode of thinking merely
by recalling explicitly the various antecedent factors which deter-
mine the quantitative value of any given construct which offends
them, each time it is encountered.

THE SYMBOLIC REPRESENTATION OP HABIT STRENGTH

From the foregoing it is evident that the chief advantage to
be expected from the employment of the logical construct habit
strength arises from economies in thought, i.e., in symbolic manip-
ulation. In order to realize this advantage an appropriate sym-
bolism must be devised. The reader will be aided in the under-
standing of this appropriateness if he will recall a few relevant
relationships presented earlier. Specifically it will be well for
him to remember that the process of reinforcement sets up a con-
nection in the nervous system whereby an afferent receptor dis-
charge (s) originally involved in a reinforcement is able to initiate
the efferent discharge (r) also involved in the reinforcement. But
since the afferent discharge (s) is initiated by the action of a
stimulus energy (S) on the receptor, and since the efferent dis-
charge (r) in due course enters the effector system, producing a
reaction (i?), we have the sequence,

S->s >r^>R.

The habit organization is represented by the arrow with broken
shaft between the neural processes s and r. If we replace this
arrow as a representation of habit with the more convenient and
somewhat more appropriate letter H, we have the full and explicit

I 12

PRINCIPLES OF BEHAVIOR

notation for expressing the various relationships involved in the
concept of habit strength:

S-*;H r -+ R.

However, under most circumstances there is a close approxima-
tion to a one-to-one correspondence, parallelism, or constancy be-
tween S and s on the one hand and between r and R on the other.
Accordingly, for purposes of coarse molar analysis <S or s may be
used interchangeably, as is the case with r and R. Since we shall
be dealing with gross stimulus situations and the gross results of
molar activity in the early stages of the present analysis, we shall
usually employ the symbol,

sH*.

Later, when we reach a point requiring a more precise and detailed
analysis, it will be necessary not only to employ the notation

i H r ,

thus explicitly representing the neural impulse, but to distinguish
through further subscript modifications various aspects of both the
stimulus and the response situations. For example, S and s repre-
sent S and s when considered as in the process of being conditioned,
whereas the dots will never be used when S and s are considered
as performing the function of response evocation.

HABIT STRENGTH CONCEIVED AS A FUNCTION OF THE NUMBER

OF REINFORCEMENTS

Having decided to employ the construct bHr, we proceed at
once to the problem of determining the presumptive quantitative
nature of its functional relationship to its various antecedent deter-
miners. The first of these to be considered will be the relationship
of sHr to the number of reinforcements ( N ). This type of deter-
mination presents certain difficulties.

Where the members of a functional relationship are both directly
measurable, as in the Hovland reaction-amplitude study cited
above, the procedure for determining the approximate mathematical
relationship is fairly straightforward. A table of corresponding
empirical values of the two variables is prepared and usually
plotted on graph paper, as in the circle sequence of Figure 21.
From an inspection of these empirical results various equations
known to yield curves resembling the one shown in the graph are

aH m AND THE NUMBER OF REINFORCEMENTS 1 1 3

fitted to the data. This consists in the main in calculating various
constants 1 called for by the respective equations. Thus in the
equation fitted to the Hovland data shown on page 120, the values
of .141, .033, and 3.1 are all fitted, or empirical, constants; the 10
is an arbitrary value chosen for convenience because it is the base
of common logarithms. In the end that equation is accepted, along
with the values of the various constants associated with it, which,
when the several values of one of the two variables are substituted
in it, yields the closest approximation to the corresponding values
of the other. In Figure 21 this approximation is shown by the
nearness of the circles to the curve which was generated by the
equation (see terminal notes).

When, on the other hand, one of the variables of a functional
relationship under investigation is a logical construct and so is
neither observable nor directly measurable, the situation is quite
otherwise, and the procedure for determining the quantitative rela-
tionship is necessarily indirect and more difficult. The procedure
in this case is to a considerable extent trial and error in nature,
though in a rather different sense than where both sets of values
are directly measurable. However, not all is trial and error, since
in both situations certain supplementary principles are usually
available for tentative guidance. This is notably true in the case
of the probability-of-reaction-evocation curve of learning (see
Chapter XVIII, p. 326 ff.).

The investigation of the functional relationship of habit strength
to the number of reinforcements is so new that the greater part of
the trial and error involved in its determination has yet to be
performed, even though the present attempt is the second such trial
to be made. Taking our point of departure from extensive obser-
vations in the field of habit formation typified by the experiments
which yielded Figures 21 to 24, and profiting by the outcome of
the first such attempt {2, pp. 164-165), it is concluded that very
probably:

1. Habit strength is an increasing function of the number of rein-
forcements.

2 . This function increases up to some sort of physiological limit be-
yond which no more increase is possible.

a A rather elaborate example of the determination of such constants in
the fitting of a curve to empirical learning data may be found in Chapter
XII, p. 200; another may be found in reference 2 , pp. 103-108. At bottom,
much of this is dependent in one way or another upon the use of simultaneous
equations.

PRINCIPLES OF BEHAVIOR

XI 4

3. As habit strength approaches this physiological limit with con-
tinued reinforcements the increment (A sHr) resulting from each addi-
tional reinforcement decreases progressively in magnitude.

Now, there are numerous algebraic expressions which yield re-
sults conforming to the above specifications. One of these, how-
ever, has a rather special promise because it is known to approxi-
mate closely a very large number of observable empirical relation-
ships in all sorts of biological situations involving growth and
decay. Indeed, Figures 21, 22, and 23 are all cases in point. The
basic principle of the simple positive growth function (Figures 21
and 23) is that the amount of growth resulting from each unit of
growth opportunity will increase the amount of whatever is grow-
ing by a constant fraction of the growth potentiality as yet un-
realized.

THEORETICAL CURVE OF HABIT- STRENGTH GROWTH

EXEMPLIFIED

The characteristics of the positive growth function may be ex-
hibited by means of an example. From the foregoing it is evident
that the rate of habit growth is dependent upon three factors or
parameters:

1. The physiological limit or maximum (M)

2. The ordinal number ( N ) of the reinforcement producing a given
increment to the habit strength (A b^r)

3. The constant factor (F) according to which a portion (A bHr)
of the unrealized potentiality is transferred to the actual habit strength
at a given reinforcement

There must also be devised a unit in which to express habit
strength. This is taken arbitrarily as 1 per cent of the physiologi-
cal maximum ( M ) of habit strength attainable by a standard
organism under optimal conditions. In order to make the name of
the unit easy to remember, it will be called the hab, 1 a shortened
form of the word habit. Thus under the conditions stated above
there would be 100 habit units, or habs, between zero and the

100

physiological limit, i.e., one hab —

r JL \J\J

We proceed now with our example. Suppose that the growth
constant (F) in a given reinforcement situation is taken as 1/10-

1 Pronounced hob , as in cab.

sHb and the number OF REINFORCEMENTS i 1 5

One-tenth of the total possibility of learning (100 units) is 10 habs
(1/10 of 100 = 10). The generation of 10 units of habit strength
from a base, zero, leaves 100 — 10, or 90 units of growth yet pos-
sible of realization. Consequently the habit increment resulting
from the second reinforcement must be 1/10 of 90, or 9; i.e., the
second A sH R = 9 habs. Subtracting 9 from 90, we have left 81
units of possible growth. One-tenth of 81 in turn yields our next
A bH r of 8.1 habs; and so on. This process can be repeated as
many times as there are successive repetitions of the reinforcement.
Column 2 of Table 1 shows the first 30 successive A com-
puted in this way. These are shown graphically at the left

TABLE 1

«,r^ NALYTICAL Table Showing the Theoretical Evolution of a Typical
Growth” Function in Which Each Increment to the Habit Is 1/10 of
the Potential Habit Strength as yet Unformed. (See text for details and
* igures 25 and 26 for graphical representation.)

Ordinal Number of
Reinforcements

Increment of Habit
(A kH R )

Total Accumulated
Habit in Hab Units
(2A sHr)

8.1

27.1

7.29

34.39

6.561

40.951

5.9049

46.856

5.3144

52.1703

4.7830

56.9533

4.3047

61.2580

3.8742

65.1322

3.4868

68.6189

3.1381

71.7570

2.8243

74.5813

2.5419

77.1232

2.2877

79.4109

2.0590

81.4698

1.8530

83.3228

1.6677

84.9905

1.5009

86.4915

1.3509

87.8423

1.2158

89.0581

1.0942

90.1523

.9848

91.1371

.8863

92.0234

.7977

92.8210

.7179

93.5389

.6461

94.1850

.5815

94.7665

.5233

95.2899

.4710

95.7609

PRINCIPLES OF BEHAVIOR

1 16

edge of Figure 25, piled one upon the other in the order in which
they were derived. It is notable that these increments become
smaller and smaller until, with very large values of N, they become
infinitesimal.

In column 3 of Table 1 are presented the cumulative values of
column 2. These latter values, in their turn, are represented graph-
ically in the main portion of Figure 25. The contour of this
columnar figure is a rather precise representation of what is here
conceived to be the basic “curve of learning” from which all other
theoretical curves of learning are derived in one way or another.

Fia. 25. Diagrammatic representation of a theoretical simple positive
growth function. At the left are given the successive increments of habit
accretion for successive reinforcements as shown in column 2 of Table 1. At
the right may be seen the amount of accumulated habit strength at the
successive reinforcements as shown in column 3 of Table 1.

It will be noticed that it rises at first with comparative rapidity,
the rate of rise gradually diminishing until at high values of N it
becomes practically horizontal. Because of their progressively
diminishing rate of rise such curves are said to be negatively accel-
erated.

The notched appearance of the contour of the main portion
of Figure 25 is due to the fact that each reinforcement is a unit,
i.e., is essentially indivisible into fractional parts such as halves
or thirds of a reinforcement. The usual method of plotting learning
functions by smooth-line curves running through the points repre-
senting the readings taken after each reinforcement, is shown for

sHi i AND THE NUMBER OF REINFORCEMENTS 1 17

the values in column 3 of Table 1 in Figure 26. It is to be noted,
however, that while for many purposes this method of representing
the course of learning is to be preferred because of its convenience,
there is danger of its giving the uninitiated a false impression of
smooth continuity. Such a smooth, continuous process could result

Fia. 26 . Theoretical curve of learning a simple conditioned reaction (bHk)
as a function of the number of reinforcements ( N ), plotted in the customary
manner, which implicitly but falsely assumes that repetitions can be sub-
divided indefinitely (see text).

only if the successive repetitions of reinforcement were to be indefi-
nitely subdivided into fractional parts. The cyclical nature of the
reinforcement process precludes this.

SUMMARY

The effect of reinforcement may become manifest in overt action
upon the presentation of the associated stimulus at any time during
the subsequent life of the organism. This central fact shows con-
clusively that reinforcement leaves within the organism a relatively
permanent connection between the receptor and the effector asso-
ciated in the original reinforcement. It is this which in the present
system is meant by the term “habit,” a technical adaptation of
the common-sense concept that goes by the same name.

Since the organization of the nervous system upon which habit-
ual action is evidently based lies deeply hidden and quite remote

PRINCIPLES OF BEHAVIOR

1 18

from any immediate means of direct observation, habit has the
status of an unobservable, i.e., it is a logical construct. As such
it is prevented from becoming a metaphysical entity by being firmly
anchored, both antecedently and consequently, to phenomena which
alike are observable and measurable. On the antecedent or causal
side, habit is known to be dependent upon various factors asso-
ciated with reinforcement, but notably upon the number of rein-
forcements. On the consequent or effect side, habit manifests itself
in action, ideally on the presentation of the stimulus aggregate
originally associated with it at its reinforcement. It is important
to note, however, that between 8 H R and observable response phe-
nomena there intervene several additional symbolic constructs (see
Figure 84), each of which is directly or indirectly anchored both
antecedently and consequently to quantitatively observable phe-
nomena. The strength of the habit is manifested indirectly by
various measurable aspects of action: (1) reaction amplitude or
magnitude ( A ), (2) reaction latency (3) resistance to ex-

perimental extinction (n), and (4) probability ( p ) of occurrence,
i.e., per cent of appropriate stimulations which evoke the associated
reaction (p. 326 ff.).

From a study of the empirical relationships of the number of
reinforcements to typical examples of each of the four forms of
habit action it is concluded that while all are dependent in the
main upon habit strength, each is also dependent in part, and dif-
ferentially so, upon other factors which enter the reinforcement
situation. For this reason it will be convenient to have a repre-
sentation of habit strength independent of any of its potential
behavioral manifestations.

The determination of the functional relationship of an observ-
able to an unobservable presents a rather different and more diffi-
cult problem than that of two observables. The two determinations
are alike in that they are both dependent partly upon a process
of trial and error, though each in a somewhat different sense. In
the case of an observable and an unobservable the relationship
which is most plausible in the light of all related observable phe-
nomena is postulated. This postulated relationship is then em-
ployed in all appropriate deductive situations. If the assumption
is false such deductions will lead, sooner or later, to inconsistencies
with observation and so to its correction. On the other hand, i
a very large number of such deductions uniformly agree with obsei

bHm and the number of reinforcements

vation, this will indicate that the postulated relationship is valid
to an increasing degree of probability.

The postulated relationship of habit strength to the number of
reinforcements is that each reinforcement results in the addition
of an increment to the habit strength (A S H R ) which is a constant
fraction (F) of the difference between the physiological maximum
(A/) of habit strength and the habit strength immediately preced-
ing the reinforcement. This is a relatively uncomplicated mathe-
matical relationship which we shall call a simple positive growth
junction.

NOTES

The Present Trial-and-Error Status of Our Hypothesis as to the Relation

of Habit Strength to Number of Reinforcements

On page 113 it was pointed out that the determination of the correct functional
relationship of an unobservable to an observable is to a considerable extent
dependent upon trial and error. As a matter of fact, the formulation of this
relationship contained in the present chapter is the second such trial. The first
such formulation made up a part of a postulate set upon which was based a highly
formalized theory of rote learning (2). The assumption in that case was that
habit strength is directly proportional to the number of reinforcements up to the
physiological limit. That postulate generated a theorem which is clearly con-
trary to fact ( 2 , pp. 164-165). The present formulation corrects the defect thus
revealed and so is presumably a closer approximation to the truth than was the
first attempt; therefore it may be expected to survive somewhat longer.

How to Compute Habit Strength

The rather clumsy method of generating the theoretical learning curve used
above for illustrative purposes (Table 1 and Figures 25 and 26) was chosen for
expository reasons because of its psychological simplicity. For systematic pur-
poses the outcome of this arithmetical procedure as shown in column 3 of
Table 1 is usually represented by an equation. It may be shown by rather sim-
ple mathematical procedures that column 3 (2 AsHr) as a function of the num-
ber of repetitions ( N) is given by the equation:

sHr = M - Me-** (1)

where M = 100, N is the number of reinforcement repetitions, e is 10, and

* = lo e T^~F (2 >

where F is the reduction constant, in the above example taken as 1/10, i.e.,
F = .1. Accordingly,

i = log = log-i = log 1.111111+

Now, by ordinary logarithm tables,

log 1.111111+ = .04574 (approximately).

I 20

PRINCIPLES OF BEHAVIOR

Therefore, if we wish to determine the amount of sHr after five reinforcements
we have,

sH r = 100 -

100

= 100 -

= 100 -
= 100 -

IQ 04574 X 6

100

10 2287

100

1.6932

59.0597

= 40.94,

which, except for the facts that logarithm tables give only approximate values
and that decimals have been dropped, would agree exactly with the corresponding
values in Table 1.

How to Compute Increment of Habit Strength per Single Reinforcement

In a similar manner the value of the increment in habit strength due to one
reinforcement (A sHr) at any particular stage of the learning is given by the
equation,

A s H r = M - x - (M - x) 10- (3)

where x is the strength of the habit immediately preceding the reinforcement
which produces the increment, and M , i, etc., have the same values as above.
For example, the value of A a Hr which would result from a sixth reinforcement
following the fifth reinforcement (the aH R of which was calculated above) is
calculated as follows: substituting in equation 3, we have,

A b H r = 100 - 40.94 - (100 - 40.94)10" 04674

= 59.06 -

59.06

10 04574

= 59.06 -

59.06

1.1111111

= 59.09 - 53.154
= 5.906.

The value 5.906 is as close an approximation to the value in column 2 of Table 1,
where N = 6, as is to be expected from the approximations attainable where
ordinary logarithm tables are employed in the computations.

Equations Fitted to the Learning Curves

We turn now to the details of the analysis of the data of Figures 21, 22, and 23.
The Hovland data represented in Figure 21 are fitted fairly well by the equation,

A = 14.1(1 - 10- o 33 *) +3.1, ( 4 >

from which the curve running through the data points of Figure 21 has been
plotted. This is a simple positive growth function.

bHm and the NUMBER OF REINFORCEMENTS 12 i

The Simley data, represented in Figure 22, are fitted fairly well by the equa-
tion,

, _ 5.11

** 1100(1 - 10 - 12 * *)]«.’ (5)

from which the curve running through the data points of Figure 22 was plotted,
where st R represents the time (0 from the beginning of 5 to the beginning of R,
i*e*» the reaction latency. The equation represents the reciprocal of a slightly
complicated positive growth function. Since the denominator of the right-hand
member of the equation becomes zero when N = 0, st R becomes infinite under
the same conditions, i.e., no reaction whatever occurs when there is no habit
strength. In point of fact, no reaction occurs when the effective reaction potential
is less than the reaction threshold (see Chapter XX).

The Will iams data represented in Figure 23 are fairly well expressed by the
equation,

n = 66(1 - 10- 018 *) - 4, (6)

from which the curve drawn through the data points of Figure 23 was plotted,
where n represents the number of unreinforced reactions performed before a
given degree of experimental extinction develops.

The Relation of the Equation Expressing Habit Strength (1) to

Equations 4, 5, and 6

It is implicit in the foregoing analysis that equations 4, 5, and 6, and the data
represented in Figure 24, are composites, one component of which in each case
takes the general form of equation 1. For example, equation 4,

A = 14.1(1 - 10" 033 ^ + 3.1,
when thus analyzed breaks up into the following:

, b H r = 100 - 100 X lO" 833 x (7)

and

A = .141 X bH r -f 3.1. (8)

Of these, equation 7 expresses habit strength (sJ7«) as a function of the number
of reinforcements, and equation 8 expresses the amplitude of the reaction as a
function of s H Rt the latter relationship being linear (see Figure 87).

The equation from the analysis of bHr as a function of N which emerges from
equation 5 (Figure 22) is,

b Ur = 100 - 100 X 10- 3 *2 ”, (9)

and that emerging from the analysis of equation 6 (Figure 23) is,

b Hr = 100 - 100 X 10-- 018 x. (10)

It will be observed that the general form of equations 7, 9, and 10 is the same,
though there is a wide variation in the coefficient of N.

The various equations corresponding to equation 8 represent the joint rela-
tionship of the final effector process to (1) the results of previous learning retained
in the nervous system and (2) the particular stimulus conditions existent at the
time of action evocation. Since numerous factors other than the mere s Hr
enter into the ultimate evocation process, the various functional relationships

122

PRINCIPLES OP BEHAVIOR

corresponding to equation 8 cannot properly be taken up until these factors have
been examined. Accordingly, the final analysis of Figures 21, 22, 23, and 24
will be delayed until Chapter XVIII.

Does s H r Qualify as a Quantitative Scientific Construct?

Even though it be granted that the nature and quantitative aspects of habit
action are immediately and necessarily dependent upon the state of the nervous
system, it does not follow that this state, as represented by s Hr, qualifies as a
satisfactory scientific construct. As already pointed out in the text, a typical
scientific construct represents the joint, unitary action of a number of independent
directly measurable variables in the determination of some subsequent event.
If these variables do not act as a unit in a given situation they cannot 'properly he
treated as a quantitative construct in that situation.

For example, there are excellent empirical grounds for believing that habit
strength is dependent not only upon the number of reinforcements, but upon a
number of other measurable antecedent conditions under which reinforcement
occurs. These antecedent factors play the r61e of independent variables. When
finally worked out, the quantitative value of b Hr is to be thought of as a de-
pendent variable whose value may be calculated by substituting in a more or
less complex equation or formula the values of numerous independent variables
such as the number of reinforcements, the magnitude of the reinforcing agent
employed, the time from the onset of the conditioned stimulus to the reaction,
the time from the reaction to the reinforcement, the number and nature of the
irrelevant stimuli present during the reinforcement, and so on. All of these
independent variables may be assumed to be related to habit strength in different
ways, some favoring it and others hindering it in varying degrees. Now, in such
a situation it is evident that an increase or decrease in the value of one of these

independent variables may exactly offset a certain amount of decrease or increase
in the value of any of the others. This means that a given amount of habit
may be produced indifferently by innnm p.ra.ble combinations of the antecedent
variable values. It means, further, that all of these different combinations of ante-
cedent reinforcement variables will yield, other factors equal , in the action evocation
situation exactly the same amplitude of reaction , latency of reaction, persistence of
reaction , and probability of reaction. Fortunately these implications of the
unitary nature of a truly scientific construct are capable of fairly straightforward

empirical test.

It is evident that whether b Hr, or indeed any other behavioral construct,
will satisfy the requirements just outlined depends upon the outcome of a great
amount of precise quantitative experimentation, only a little of which has yet
been performed. This uncertainty applies not only to the constructs employed
in the present system but to those of all other theoretical approaches, including
the potential systems of the various Gestalt schools. If the ultimate verdic
proves to be affirmative, the task of the behavior sciences will be far simpler
than if it turns out to be negative. Meanwhile we must resolutely face the reali-
ties of the situation, whatever the future holds for us. It is believed that t e
quickest and most economical way to discover whether or not behavior is so
constituted as to permit the use of genuine scientific constructs is boldly to pos u-
late it, deduce the implications of this assumption in all possible situations, an
then accept or reject the hypothesis as the deductions agree or disagree

bHb and the NUMBER OF REINFORCEMENTS 123

empirical findings. The beginning of such an attempt is being made in the
present work.

REFERENCES

1. Hovland, C. I. The generalization of conditioned responses. IV. The

effects of varying amounts of reinforcement upon the degree of generali-
zation of conditioned responses. J. Erper. Psychol., 1937, 21, 261-276.

2. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. M athematico -deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

3. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London: Ox-

ford Univ. Press, 1927.

4. Perin, C. T. Behavior potentiality as a joint function of the amount of

training and the degree of hunger at the time of extinction. J. Exver
Psychol, 1942, SO, 93-113.

5. Simley, O. A. The relation of subliminal to supraliminal learning. Arch.

of Psychol, 1933, No. 146.

6 . Williams, S. B. Resistance to extinction as a function of the number of

reinforcements. J. Exper. Psychol., 1938, 28, 506-521.

7. Woodrow, H. The problem of general quantitative laws in psychology.

Psychol Bull., 194 2, 39, 1-27.

CHAPTER IX

Habit Strength as a Function of the Nature and
Amount of the Reinforcing Agent

We have seen that if a habit is to be set up, the act involved
must be associated either with a need reduction or with some
stimulus which has itself been associated with a need reduction.
Now, the amount or quality of the reinforcing agent at every rein-
forcement may clearly vary in such a way that the degree of need
reduction will range from very large amounts through very small
amounts down to a value of zero, at which point presumably no
reinforcement at all will occur. In short, the amount (or quality)
of the agent employed at each reinforcement appears as the second
of the numerous antecedent conditions determining habit strength.

It is evident even from the preceding analysis that somewhere
between the extremes of zero need reduction and a maximum value
of such reduction a transition must be made from zero amount of
the reinforcing agent to an amount of considerable magnitude. The
question arises: Is this transition abrupt, or is it gradual and pro-
gressive? And if it is progressive, what is the law of its progress?
A small amount of experimental evidence bearing on this question
is available, some of it concerned with conditioned-reflex learning
and some with selective learning.

THE LIMIT OF CONDITIONED-REFLEX LEARNING AS A FUNCTION

OF THE AMOUNT OF THE REINFORCING AGENT

Gantt ( 2 ) has reported a conditioned-reflex experiment of the
Pavlovian type in which each of several animals was conditioned
to four different stimuli, one stimulus being reinforced by one-half
gram of food, one by one gram, one by two grams, and one by
twelve grams. The four conditioned reactions were reinforced in
random order not only on different days but during the experimen-
tal session on the same day. After considerable amounts of train-
ing it was found that some dogs developed clearly differentiated
reactions to the several stimuli, though others were unable to do
this. The mean results from one of the former, named “Billy,”
a very stable animal, are shown by the circles in Figure 27. The

124

125

sHh and the reinforcing agent

curve running through these values is a simple positive growth
function originally fitted to them by Dr. Gantt ( 2 ). The close-
ness of the approximation of the fitted to the empirical values indi-
cates considerable consistency despite the small number of points
involved.

Notwithstanding the resemblance of this curve to those char-
acteristic of ordinary learning, it is not to be confused with a

Fro. 27. Graphic representation of the empirical functional relationship
between the amount of the reinforcing agent (food) employed at each rein-
forcement of four conditioned reactions to as many different stimuli, and the
final mean amount of salivary secretion evoked by each stimulus at the
limit of training. The appreciable secretional value of 75 units when the
fitted curve is extrapolated to where the amount of reinforcing agent equals
zero is presumably due to secretion evoked by static stimuli arising from the
experimental environment. Plotted from unpublished data from the dog
“Billy,” kindly furnished by Gantt (2) and here published with his permis-
sion. The experimental work upon which this graph is based was performed
previous to 1936 (personal communication from Dr. Gantt).

learning curve; on the contrary, each of the data points of this
investigation represents the mean response on the part of the dog
at the limit of training to the respective amounts of reinforce-
ment; i.e., each represents the final horizontal portion or asymptote
of a separate and distinct curve of learning.

THE RATE OF SELECTIVE LEARNING AS A FUNCTION OF THE

AMOUNT OF THE REINFORCING AGENT

Grindley (3) has reported a study which is closely analogous
to that of Gantt but which involves selective rather than condi-

126

PRINCIPLES OF BEHAVIOR

tioned-reflex learning. This investigator trained five groups of
twelve-day-old chicks to traverse a runway eight inches wide and
four feet long. The reinforcing agent or reward was grains of
boiled rice placed in a shallow tray at the end of the runway. One
group of chicks found and ate one grain of rice on reaching the
tray after successfully traversing the runway; one group received
two grains; another, four grains; and a fourth, six grains. A fifth
group received no food whatever on reaching the tray. The score

adopted as an index of learning was i.e., 100 times the recipro-

cal of the time in seconds required by a chick to traverse the
runway and begin eating the rice.

Grindley published composite learning curves for his several
groups of chicks. Pooled measurements of the comparable later

Fig. 28. Empirical graph representing the rate of selective learning as a
function of the magnitude of the reinforcing reward. Each circle represents
the mean score at the last five of seven trials of ten chicks in traversing a
four-foot runway to secure differing numbers of boiled-rice grains. (Derived
from measurements made of learning curves published by Grindley, 3 , p. 174.)

portions of these several curves are shown in Figure 28. In order
to permit further comparison with the Gantt study, a simple posi-
tive growth function has been fitted to the Grindley data; this is

sHa and the reinforcing agent

I2 7

represented by the curve passing among the circles of Figure 28.
While the deviations of the data points from the curve are con-
siderable, the fit is still close enough to indicate a fair approxima-
tion.

Grindley’s results are corroborated and extended by a study
reported by Wolfe and Kaplon (4). These investigators repeated
the substance of Grindley’s experiment, utilizing with groups of
chicks an explicit trial-and-error task, that of learning a simple
T-maze. A comparison was made between the reinforcing power
of one-fourth of a kernel of popcorn, a whole kernel, and four
quarters of a kernel given at once. It was found that a whole
kernel was more reinforcing than a quarter of a kernel, which
confirms the findings of both Gantt and Grindley. However, four
quarter kernels given at once proved to be distinctly more rein-
forcing than did a single intact kernel. In the case of the four
quarter kernels, of course, the chickens pecked and swallowed four
times after each successful act. We thus appear to have in this
case the paradox of four more or less distinct reinforcements con-
tributing by summation to produce the increment of learning re-
sulting from a single successful action sequence.

In spite of the great difference in the organisms involved, in
the index of learning employed, and in the stage of learning repre-
sented, the results of the selective learning experiments show a strik-
ing general agreement with the conditioned-reflex results of Gantt.
It accordingly seems fairly clear from the two types of studies that
the rate of learning is an increasing monotonic function of the
amount of the agent employed at each reinforcement.

HOW DOES THE AMOUNT OF THE REINFORCING AGENT INFLUENCE
THE TWO PARAMETERS OF THE CURVE OF HABIT FORMATION?

We saw above (Chapter VIII) that the curve of habit strength
as a function of the number of reinforcements was dependent upon
two constants, (1) the physiological maximum of habit strength
( M ), and (2) the fractional part ( F ) of the as yet unrealized
potentiality of habit-strength acquisition, which is added to the
actual habit strength at each reinforcement. Assuming the sound-
ness of this hypothesis, it is evident that the influence of increas-
ing the amount of the reinforcing agent on the size of the increment
of habit strength (A bHr) at a given reinforcement must result
from an increase in one, and possibly both, of these parameters*

128

PRINCIPLES OF BEHAVIOR

The parameter, or parameters, involved could be determined if
we had reliable learning curves which were carried up to, or near,
the limit of practice under different amounts per reinforcement of
the same reinforcing agent. Unfortunately no such investigation
has yet been reported in detail, though there are indications in
both the study of Gantt and in that of Wolfe and Kaplon that,
at the limit of practice, habit strength is definitely greater where
the amount of the agent employed in reinforcement is greater.
Gantt has published no practice curves, and those published by
Wolfe and Kaplon are not sufficiently regular to make profitable
a precise analysis from this point of view. An inspection of these
latter curves and the tables accompanying them suggests, however,
that in habit formation the F - value may be approximately constant
for different amounts of the reinforcing agent or reward. The only
possible remaining parameter which could produce the slower learn-
ing with small amounts of the reinforcing agent is the asymptote

or upper limit of the learning curve.

In this connection it will be recalled from the last chapter that
the physiological limit of habit strength under absolutely optimal
conditions was taken as 100 habs, p- 114. To this value was
assigned the symbol M. The introduction of the presumption that
the asymptotes of learning curves may vary below this level, de-
pending on the amount and quality of the reinforcing agent, makes
it necessary to employ a separate symbol (Af') to represent such
limits or asymptotes. We now state the working hypothesis at
which we have arrived: In a learning situation which is optimal
in all other respects, the limit ( M ') of habit strength ( gH R ) attain-
able with unlimited number of reinforcements is a positive growth
function of the magnitude of the agent employed in the reinforce-
ment process. This tentative conclusion is based on admittedly

inadequate grounds and will therefore be subject to reexamination
and revision when more satisfactory evidence becomes available.

An important extension of this hypothesis at once suggests
itself. Clearly a reinforcing agent may vary in quality as well as
in quantity. More specifically, with quantity remaining constant,
a reinforcement by one agent may reduce the need more than a
reinforcement by another. For example, a standard food may be
adulterated by adding varying amounts of some inert and tasteless
substances such as a flour consisting of ground wood. It is evident
that two grams of a standard dog ration mixed with 50 per cent
wood flour would reduce an animal’s food need only half as muc

bHb AND THE REINFORCING AGENT 129

as would the same quality of the food unadulterated. While no
report of such an experiment has been found, it seems reasonable
to suppose on the analogy of the experiments of Gantt and Grind-
ley that a food so adulterated would be less effective as a “pri-
mary” reinforcing agent than would an equal weight of the natural
food. However, judging from the shape of the curve of Figures 27
and 28 it is to be expected that the limit of learning (m) resulting
from reinforcement with the adulterated agent would be more than
half as effective as would be that resulting from the use of the
unadulterated agent.

We conclude, then, that habit strength at the limit of practice
(m) will vary with the quality , as well as the quantity, of the rein-
forcing agent from a minimum of zero to a physiological maximum
of 100 habs, and that the rate of approach to that limit (F) will
remain unchanged.

SOME IMPLICATIONS OF THE AMOUNT-OF-REINFORCEMENT

HYPOTHESIS

The meaning of the working hypothesis just formulated may be
clarified by indicating one or two of its implications. Let it be
supposed, on the analogy of the Gantt experiment, that the maxi-
mum amount of a given food when used as a reinforcing agent would
yield at the limit of practice a habit strength of 80 habs. Calcula-
tions based on the F-constant fitted to the Gantt data show that if,
under these conditions, one gram of this food were used at each
reinforcement, the maximum habit strength to be expected at the
limit of practice would be 23.75 habs. A parallel computation
shows that the maximum to be expected from the use of six grams
of this food at each reinforcement would be 70.14 habs.

With these maxima available it is possible to calculate the theo-
retical course of habit-strength acquisition under the respective
conditions by substituting first one of the values for m in the simple
positive growth function, and then the other, letting the fractional
incremental factor, F, equal 1/10, as in the illustration of Chapter
VIII. In this way were computed the values from which were
plotted the two main curves of Figure 29. These curves, taken
together, show the effect upon the course of habit formation implied
by the hypothesis put forward above.

An additional implication of the working hypothesis may still
further clarify its meaning. Let it be supposed that the habit

130 PRINCIPLES OF BEHAVIOR

has been reinforced with one gram of the food fifteen times and
that the reinforcement is then suddenly shifted to six grams on the
next fifteen reinforcements. Neglecting the presumptive persever-
ating influence of secondary reinforcement in the situation, the out-
come is easily calculated by methods analogous to the determina-
tion of the two main curves of Figure 29. This is shown by the
dotted curve rising from the one-gram curve at the fifteenth rein-

Fia. 29 . Graphic representation of the theoretical course of habit-streng
acquisition with a six-gram food reinforcement (broken line) and with a ° ne ”
gram food reinforcement (solid line). The dotted curve indicates the theo-
retical course of habit-strength acquisition on the assumption that re }?~
forcement is abruptly shifted to six grams on the sixteenth trial of t
one-gram reinforcement curve.

forcement. A glance at Figure 29 shows that according to the
present hypothesis an increase in the amount of the reinforcing
agent should be followed by a marked increase in the rate of habit-
strength acquisition; this, however, would gradually lessen as t e
new limit is approached, in accordance with the nature of t e
simple growth function. In concluding our discussion of Figure
it is to be noted that exactly analogous curves would also be pro-
duced by suitable variations in the quality of the reinforcing agen •

sHs AND THE REINFORCING AGENT

* 3 *

Finally, an additional case may be mentioned — that in which
the six-gram habit would be reinforced ten times. This would
generate a habit strength of 45.68 habs, which is considerably above
23.75 habs, the maximum attainable by means of a one-gram rein-
forcement. Then the reward would suddenly be shifted to one
gram. Assuming these conditions, and ignoring secondary reinforc-
ing effects, it is to be expected that successive reinforcements would
result in a progressive weakening of the habit. Consideration of
this interesting but complex problem must be deferred until the
phenomena of experimental extinction have been taken up in detail
(p. 258 ff.).

THE AMOUNT OF THE REINFORCING AGENT AND THE PROBLEM

OF INCENTIVE

Although the systematic presentation of the subject of moti-
vation has been reserved for a later chapter (p. 226 ff.), it becomes
desirable here to touch briefly on one of its phases. Motivation has
two aspects, (1) that of drive ( D , or S D ) characteristic of primary
needs, and (2) that of incentive. The amount-of-reinforcement
hypothesis is closely related to the second of these aspects. The
concept of incentive in behavior theory corresponds roughly to the
common-sense notion of reward. More technically, the incentive
is that substance or commodity in the environment which satisfies
a need, i.e., which reduces a drive.

Let us suppose that in a simple selective learning situation in-
volving a hunger drive, the food employed as a reinforcing agent
is plainly visible at the moment the organism performs the several
acts originally evoked by the stimulus situation; that the indi-
vidual reinforcements are separated by a number of hours; and
that the amount of food employed in the several reinforcements
varies at random from almost zero to very large amounts. Under
such circumstances it follows from the principle of reinforcement
that the visual stimulus arising from the food will be conditioned
to the successful act by the subsequent reinforcing state of affairs,
the consumption and absorption of the food. Moreover, in case the
amount of food shrinks to zero there will be no direct reinforce-
ment at all. From these considerations, coupled with the amount-
of-reinforcement hypothesis, it may be inferred that the successful
reaction will be more strongly conditioned to the stimulus aggregate
arising from a large piece of food than to that from a small one.

PRINCIPLES OF BEHAVIOR

1 3 2

Therefore, given a normal hunger drive, the organism will execute
the correct one of several acts originally evoked by the situation
more promptly, more vigorously, more certainly, and more persist-
ently when a large amount of food is stimulating its receptors than
when they are stimulated by a small amount.

Substantial confirmation of this deduction is furnished by a
recent experiment reported by Fletcher (I), who trained a chim-
panzee to secure pieces of banana by pulling into its cage a weighted
car by means of an attached rope. It was found that the animal
would perform more work for a large piece of banana than for a
small one. The maximum amount of work which would be per-
formed for a given amount of banana incentive between the range
of .64 and 3.77 units was practically linear.

An extrapolation of Fletcher’s linear relationship just men-
tioned suggests that the animals would have performed with a zero
amount of incentive more than half as much work as was performed
with the incentive at 3.77 units. This might well occur on a few
occasions due to the conditioning of the reaction to the stimuli
arising from the apparatus and other environmental elements. On
the other hand, in case no food is present the strength of the re-
maining stimulus elements conditioned to the reaction may be bo
far depleted by the absence of the incentive component of the
stimulus compound (see Chapter XV) that the effective habit
strength will either be less than the reaction threshold or than
the strength of some competing reaction tendency evoked by the
stimulus situation; in either case the reinforced reaction may not
occur at all. In the event that it does occur under such conditions,
however, experimental extinction will presently set in and soon
terminate it.

SUMMARY

Since the amount of need reduction presumably varies with the
amount of the reinforcing agent consumed by the organism, it fol-
lows as a strong probability from the dependence of reinforcement
upon the amount of need reduction that the increment of habit
strength (A gH R ) per reinforcement will be an increasing function
of the amount of the reinforcing agent employed. This a prion
expectation is substantiated by empirical investigations involving
both selective and conditioned-reflex learning. Moreover, t ese
studies, indicate that the relationship is that of a simp e posi
growth function.

s h b and THE REINFORCING AGENT 133

Because the law of simple habit acquisition is presumably a
positive growth function of the number of reinforcements, it fol-
lows that the increased rate of habit acquisition with increased
amounts of the reinforcing agent employed may be due to one or
both of two factors: (1) an increased limit of potential habit-
strength growth, or (2) an increased fraction of this potentiality
which is added to the habit at each reinforcement. The empirical
evidence on this point is at present inadequate for final decision.
Pending the appearance of more complete evidence, the working
hypothesis is adopted that an increase in either the quality or the
quantity of a reinforcing agent increases the rate of learning by
raising the limit (m) to which the curve of habit strength ap-
proaches as an asymptote, the rate of approach (F) to this limit
possibly remaining constant for all qualities and amounts of the
reinforcing agent employed.

From the amount-of-reinforcement hypothesis may be derived
a special case of one phase of motivation, that of incentive or sec-
ondary motivation. This is the situation where the incentive (rein-
forcing agent) contributes a prominent, direct component of the
stimulus complex which is conditioned to the act being reinforced.
The stimulus component arising from a large amount of this
substance will be different from that arising from a small
amount, and will differ still more from a stimulus situation con-
taining a zero amount. It follows from this and the amount-of-
reinforcement hypothesis that in the course of reinforcement by
differing amounts of the reinforcing agent, the organism will in-
evitably build up stronger reaction tendencies to the stimulus aris-
ing from large amounts than to that from small amounts, and
no habit strength at all will be generated by zero amounts. It
thus comes about, primary motivation (e.g., hunger) remaining
constant, that large amounts of the agent will evoke more rapid,
more vigorous, more persistent, and more certain reactions than
will small or zero amounts. Thus a reinforcing agent as a stimulus
becomes an incentive to action, and large amounts of the agent
become more of an incentive than small amounts. This a priori
expectation is well substantiated by quantitative experiment as well
as by general observation.

*34

PRINCIPLES OF BEHAVIOR

NOTES

The Equations Which Were Fitted to Gantt's and Grindley’s Empirical

Data

The equation fitted by Gantt (2) to the data represented by the circles shown
in Figure 27 is,

A = 335 (1 - 10- i M «) + 75, (11)

where A is the amount of salivary secretion in arbitrary units during a constant
number of seconds, and w is the weight of the reinforcing agent in grams employed
at each trial. The + 75 is arrived at by extrapolation.

The equation fitted to the Grindley data represented by the circles of Figure 28
is,

1 AA

~ = 21.38 (1 - lO-^ n') _|_ m5>

where / is the time in seconds required to traverse the four-foot runway to the
food and n (number of rice grains) is the magnitude of the reinforcing agent
employed. The 4- .5 represents the score resulting from spontaneous exploratory
activity previous to receiving reinforcement at the end of the runway.

The Equations From Which the Curves of Figure 29 Were Derived

The equation from which the upper curve of Figure 29 was derived is,

a H R = 70.14(1 - io--o«76*) f

in which the 70.14 was derived from the equation,

AT = 80(1 - 10“«* «), (12)

where w = 6 grams.

The equation from which the lower curve of Figure 29 was derived is,

b H r = 23.75(1 - 10" °«76

in which the 23.75 was derived from the equation,

Af' = 80(1 - 10- iw *),

where io = 1 gram.

The equation from which the dotted curve rising from the 1-gram curve was
derived is

bHr = (70.14 - 18.86) (1 - 10- 04676 *) -f 18.86.

REFERENCES

1. Fletcher, F. M. Effects of quantitative variation of food-incentive on

the performance of physical work by chimpanzees. Comp. Psychol.
Monogr., 1940, 16, No. 82.

2. Gantt, W. H. The nervous secretion of saliva: The relation of the con-

ditioned reflex to the intensity of the unconditioned stimulus. Proc.
Amer. Physiol. Soc., Amer. J. Physiol., 1938, 123, p. 74.

3. Grindley, G. C. Experiments on the influence of the amount of r ®^ ar

on learning in young chickens. Brit. J. Psychol , 1929-30, 20, 173-180.

4. Wolfe, J. B., and Kaplon, M. D. Effect of amount of reward and con-

summative activity on learning in chickens. J. Comp. Psychol., It* »
31, 353-361.

CHAPTER X

Habit Strength and the Time Interval Separating

Reaction from Reinforcement

We have seen in the last two chapters that habit strength is
dependent upon two measurable aspects of reinforcement: (1) the
number of reinforcements and (2) the intensity (amount and
quality) of the reinforcing agent. In the present chapter we shall
consider the functional relationship of habit strength to a third
measurable aspect of reinforcement — that of the time interval sepa-
rating the reaction being conditioned from the reinforcing state of
affairs (p. 80).

ORIGIN AND FRACTIONATION OF THE PROBLEM

The notion that the habit strength resulting from the conjunc-
tion of a receptor and an effector process is a function of the tem-
poral nearness of a reinforcing state of affairs has long been cur-
rent. It seems first to have been formulated by E. L. Thorndike
in 1913, on the basis of general observation as an aspect of his
“law of effect” (11, p. 173). Substantially the same idea was put
forward, apparently independently, by Margaret Washburn in 1926
(14, p. 335). Thorndike expresses no opinion concerning the logical
status of the principle. Washburn, however, clearly regarded it as
a secondary, rather than a primary, principle. She believed it
could be derived from the associative “law of recency,” a supposed
basic principle of learning once much in vogue but, in the light of
recent work, now regarded primarily as a function of perseverative
stimulus traces (p. 71).

Thorndike’s original hypothesis breaks up into a number of
distinguishable aspects which present convenient points of de-
parture for a systematic consideration of the subject:

1. Is there, in fact, a functional dependence of habit strength upon
the time interval separating the receptor-effector conjunction 8 C r from the
reinforcing state of affairs?

2. Assuming such a functional dependence, what is the direction of
slope of the gradient?

135

PRINCIPLES OF BEHAVIOR

136

3. How far does this gradient extend from the point of reinforce-
ment before falling approximately to zero?

4. What is the shape of this gradient; i.e., what are its mathematical
characteristics?

5. What parameters of the curve of learning are affected by the
gradient of reinforcement — the rate of rise (F), or the limit of rise (m), or
both?

6. Is the relationship a primary principle or is it a secondary one;
i.e., can it be derived from other and more basic principles?

7. What other behavior processes, if any, are derivable as secondary
phenomena from this principle?

EARLY DIRECT EXPERIMENTAL ATTACKS ON THE PROBLEM

A great deal of experimental effort has been devoted to the
solution of one or another of the subsidiary problems growing out
of Thorndike’s original hypothesis. The first of these studies, pub-
lished in 1917, was by John B. Watson (15). The apparatus of
this experiment consisted essentially of a food chamber surrounded
by sawdust to a depth of four inches. The task of the subjects,
twelve hungry albino rats, was to dig through this sawdust, find
a round hole giving access to the food chamber, and secure food
which was in a shallow cup covered by a lid. Perforations in the
lid allowed free passage of food odors. As preliminary training,
the animals lived for a time in the food chamber, where they appar-
ently ate freely from the food cup.

When the digging tests began the rats were divided into two
groups, the animals of one group being allowed to eat as soon as
they reached the food cup; but when the animals of the second
group reached the food cup the perforated cover was field in place
for 30 seconds before eating was permitted. In the course of 27
trials both groups of animals gradually reduced the time of reach-
ing the food cup from around 100 seconds to six or seven seconds,
though there were no indications of any special advantage in rate
of learning of either group over the other.

The next experiment to throw much light on the problem was
reported in 1929 by Mrs. Hamilton (nee Haas) (5). She employed
albino rats in a Warden compound Y-maze involving five succes-
sive choices, each of one correct or one incorrect turn. Between
the last choice point of the maze and the food box a retention
chamber was placed where the animals could be held as long as
desired before being permitted to enter the food box and eat. Five

bHb and the delay in reinforcement

1 37

groups of approximately 20 animals each were used, the delays in
the retention chamber employed with the respective groups being
0, 1, 3, 5, and 7 minutes.

A clearly marked difference in learning rate was found between
the group permitted to eat at once and the various delay groups,
but there was little indication of a consistent advantage in the
shorter delay groups over those subjected to longer delays. All
of the animals learned, the several delay groups requiring roughly

Fia. 30. Empirical delay-of-reinforcement gradient. The circles represent
amount of learning for constant numbers of reinforcements as a function of
the delay in the occurrence of the reinforcement. These values have been
calculated from data published by Wolfe U6).

twice as many trials to reach a given criterion of learning as did
the no-delay group. The Hamilton study, in contrast to that of
Watson and an intervening experiment by Warden and Haas { 13 ),
accordingly indicates that (1) there is in fact a gradient of rein-
forcement, (2) the gradient slopes downward from immediate rein-
forcement as the length of delay increases (both quite as Thorn-
dike and Washburn supposed), and (3) there is a suggestion that
the gradient changes its nature in some important sense when de-
lays of reinforcement exceed one minute. There is also an indica-

i 3 8

PRINCIPLES OF BEHAVIOR

tion that the employment of a separate chamber for the restraint
of the animal before it enters the food box is somehow critical in
bringing out the influence of the period of delay on learning
rate.

The next experimental investigation of the problem which de-
mands our consideration was performed by Wolfe {16). The more
significant portion of the Wolfe study concerns the learning of a
single unit of a simple T-maze by eight groups, each of eight
albino rats, with the following food delays for the respective
groups:

0" 5" 30" V 2.5' 5' 10' 20'.

Wolfe employed special delay chambers, one for the final correct
choice and one for the final incorrect choice, both being distinct

from the food chambers. In-
dices of habit strength calcu-
lated from Wolfe’s published
tables yield the values shown
graphically in Figure 30.
There it may be seen that the
gradient falls sharply from
delays of zero to those of one
minute, after which the slope,
though upon the whole con-
tinuous, is much more grad-
ual. This study is in general
agreement with Hamilton’s
findings in showing a critical
change at a delay of around
one minute; it also fills in the
important gap lying between
zero and 60 seconds by sup-
plying evidence of the rate
of learning at delays in re-
inforcement of five and 30
seconds respectively. The values in this region when carefully
examined turn out to approximate very closely a negative growth
function, as shown in Figure 31. This gives us the first convincing
clue concerning the answer to our third and fourth questions formu-
lated above.

Fio. 31. The first four of the Wolfe
delay-of-reinforcement values, together
with the negative growth or decay func-
tion which fits them rather well. This
curve represents a decrease of between
1/15 and 1/16 at each additional second
of delay in reinforcement.

bHm and the delay in reinforcement

1 39

perin’s experiment

The most recent experiment in this field is reported by Perin
(10). In that portion of his investigation which especially con-
cerns us here, Perin employed a modified form of the Skinner box
(see p. 87 above). Through a horizontal slot in a metal plate
on one wall of the experimental chamber there projected an easily

Fi a 32. Graphic representation of the gradient of reinforcement as in-
dicated by the slopes (tangents) of the composite learning curves of five of
Perms groups of animals at the point where 50 per cent of the trials were
without error. The slope of each group is represented by a circle; the
curved line running through the circles is a special form of negative growth
function which was fitted to these values. (Reproduced from Perin, 10.)

moved brass rod. During the habituation period the apparatus
was so set that a movement of this rod a few millimeters to either
the right or the left would immediately deliver a pellet of food to
the food cup beneath. After the animal had learned to perform
both acts with some facility and his preference for one of them
had been determined, the setting of the apparatus was so changed
that (1) a movement of the rod in the preferred direction gave no
food, and (2) a movement of the rod in the non-preferred direction
caused (a) the rod instantly and silently to be withdrawn and

140

PRINCIPLES OF BEHAVIOR

(6) the delivery of a pellet of food after time intervals varying
according to the group involved, as follows:

As a rule the animals sat quietly by the food cup after the with-
drawal of the bar, and simply waited for the food to be delivered.
It is not without significance that many of the animals of the
30-second group ceased to operate the bar after varying numbers
of trials; because of this the results of these animals could not be
employed in the plotting of Perm’s gradient of reinforcement.

The results of this experiment which are of special interest in
the present context are shown graphically in Figure 32. There it
may be seen that the rates of learning, as indicated by the slopes
of the learning curves of the several groups of animals at the 50 per
cent level of correctness, show a negatively accelerated descending
gradient much as does Wolfe’s study (Figure 31). There is, how-
ever, this difference: the slope of Perm’s curve is of such a nature
that, when extrapolated, it falls to zero at 34 seconds; with a some-
what different method of plotting the learning curve from which
the tangents were taken, the extrapolated gradient falls to zero
at 44 seconds.

THE RECONCILIATION OF SOME EXPERIMENTAL PARADOXES

The most outstanding paradox encountered among the experi-
mental results outlined above is the fact that the Watson study
and a comparable one by Warden and Haas ( 13 ) show no gradient
of reinforcement, whereas the Hamilton investigation and, particu-
larly, that of Wolfe clearly indicate such a gradient. As already sug-
gested, the difference in the outcomes of the two groups of studies
is probably to be attributed mainly to the fact that in the first
two the retention or delay occurred in the food chamber, whereas
in the latter two it took place in a separate (non-feeding) com-
partment.

Actually, on the basis of secondary reinforcement, we should
expect exactly such a difference in the outcome of the two groups
of experiments. It may be recalled (Chapter VII, p. 97) that
any stimulus which is closely and consistently associated with a
reinforcing state of affairs will itself gradually acquire the power
of secondary reinforcement regardless of whether the transmitting
stimulus is primarily or secondarily reinforcing. Thus in the Wat-

bHb and the delay IN REINFORCEMENT 14 1

son experiment and also in the Warden and Haas study the food
chamber would largely have acquired this power from the prelimi-
nary or habituation training for both groups of animals alike. By
the same principle, the odor of the food which came through the
perforations in the food-cup lid in both experiments had already
acquired secondary reinforcing power as the result of the animals’
having eaten the food, even before habituation to the food chamber.
There was therefore, on two counts, no delay in the effective (sec-
ondary) reinforcement in either group and so, naturally, it is not
to be expected that a delay of 30 seconds in the actual eating
would produce a retardation in the learning rate large enough to
be detected by the use of small groups of animals.

Turning, now, to the Hamilton and Wolfe studies we still find,
even with the separate retention chamber, the conditions which are
both necessary and sufficient for secondary reinforcement, though
in an attenuated form. In these investigations the eating of the
food is immediately associated with the food cup and the food
chamber by both groups alike, so that all parts of the food cham-
ber as stimuli must gradually acquire secondary reinforcing powers
as the trials increase in number. Next, since the stimuli arising
from the door giving access to the food chamber are associated
closely with the food chamber itself, this door must gradually
acquire secondary reinforcing powers following the acquisition of
these powers by the food chamber as a whole. In a similar manner,
the entire retention chamber would gradually acquire a measure
of secondary reinforcing power from reinforced association with
this door, though presumably the longer the delay there, together
with the incidental irrelevant activity during the periods of delay,
the slower would be this process. This probably explains the fact
that Hamilton’s animals continued to learn with fair speed under

delays of reinforcement up to seven minutes, and Wolfe’s animals,
under delays up to twenty minutes. *

Moreover, considering the simplicity of Wolfe’s maze, which
involved only a single choice, his animals learned notably more
slowly than did Hamilton’s, which were required to learn five
choices. This paradox quite probably was due in part to the fact
that Wolfe employed a retention chamber on the false choices as
well as on the correct ones. Presumably the two retention cham-
bers were physically identical. Under these conditions the secon-
dary reinforcing power acquired by the retention chamber next to
the food box would generalize (p. 183) to the one outside the non-

I 4 2

PRINCIPLES OF BEHAVIOR

reward box, and so tend at first to reinforce the false choice and
thereby retard the learning. In addition, the extinction effects aris-
ing from the non-reinforcement of the incorrect final choices would
generalize (p. 264) to the correct choice to some extent, thereby
weakening that reaction and still further retarding the learning.
Also presumably contributing to the relative slowness of the learn-
ing of Wolfe’s animals is the fact that inside the retention chamber
on the non-reward arm of the maze was a dish containing food
from which the animal was excluded by a wire-screen cover. In
all probability the odors and other associated stimuli arising from
this non-eating food situation constituted powerful secondary rein-
forcing influences, as in the Grindley experiment (p. 126), which

also would tend to reinforce the false choice until extinction should
supervene.

Finally, there is reason to believe that in all of the above
studies, but especially in the Wolfe study, there enters the compli-
cating factor of spatial orientation. This is the capacity possessed
by most organisms to return to a point of reinforcement if the dis-
tance is not too great. The mechanisms mediating this behavior
are complex and cannot be taken up in this place.

The conditions of Perm's experiment, on the other hand, were
designed in such a way as to preclude all irrelevant secondary rein-
forcement as well as all irrelevant spatial orientation. This pre-
sumably accounts for the fact that (1) the extrapolation of Perm’s
gradient falls to zero, whereas the extrapolation of Wolfe’s gradient
lacked much of doing this; and (2) Perm’s gradient reaches zero
at a delay of between 30 and 40 seconds, whereas Wolfe’s gradient
continues to fall for 20 minutes, and the asymptote of this fall lacks
much of being zero.

FORMULATION OF THE GOAL GRADIENT HYPOTHESIS

As we have just seen, Perm's investigation suggests that, uncom-
plicated by irrelevant factors, the basic temporal gradient of habit
strength as a function of the delay in reinforcement in the case
of the albino rat actually extends over a relatively short period of
time, possibly no more than 30 seconds and very probably less than
60 seconds. On the other hand, the Wolfe study indicates that
under ordinary learning conditions, where plenty of opportunity
for secondary reinforcement usually exists, the gradient may extend
in considerable force for a relatively long period. This means that

bHe and THE DELAY IN REINFORCEMENT 143

what was originally regarded as a single principle has turned out
upon intensive investigation to involve two fairly distinct prin-
ciples: (1) the short gradient reported by Perin, which will be
called the gradient of reinforcement , an expression coined by Miller
and Miles ( 9 ) ; and (2) the more extended gradient which is pre-
sumably generated as a secondary phenomenon from Perin’s gradi-
ent of reinforcement acting in conjunction with the principle of
secondary reinforcement. This second and more extended gradient
may with some propriety retain the original name of the goal
gradient , an expression employed by the present author in his first
discussion of the subject ( 6 ).

Unfortunately it is not yet clear in exactly what quantitative
manner the basic gradient of reinforcement, in combination with
secondary reinforcement, generates the more extended goal gradient.
We do, however, have a number of promising leads. Because of
the intimate relation known to exist between the conditioning of
a stimulus to a reaction and the acquisition by that stimulus of
secondary reinforcing power (p. 100), it is plausible to assume
that stimuli acquire this power according to the primitive gradient
of reinforcement demonstrated by Perin. It is also assumed that
once a stimulus has acquired a certain amount of the capacity for
secondary reinforcement, this will immediately begin to operate
according to Perin’s primitive gradient of reinforcement to rein-
force antecedent receptor-effector connections and to endow each
newly associated receptor process with secondary reinforcing
powers, and so on. Thus the goal gradient would result from the
summation of an exceedingly complex series of overlapping gradi-
ents of reinforcement, in part consisting of, but largely derived
from, the “primary” reinforcement occurring at the end of the
temporal period covering the behavior sequence involved. Also, the
generation of overlapping secondary gradients presumably would
take place under ordinary learning conditions not only beyond the
range of Perin’s primitive gradient of reinforcement, but within its
range as well, so that the period preceding “primary” reinforcement
by 30 or 40 seconds would present a picture not very different from
other portions of the total range, though Wolfe’s results (Figures
30 and 31) suggest that the characteristics of the goal gradient for
the first minute of the delay in reinforcement may differ somewhat
from the more remote portions. In view of the all but universal
prevalence of the conditions which generate secondary reinforce-
ment, coupled with the great difficulty experienced by Perin in

>44

PRINCIPLES OF BEHAVIOR

eliminating them from his experiment, it is fairly evident that the
principle immediately concerned in ordinary learning situations is
what we have called the goal gradient, whereas both the gradient

TABLE 2

This Table Shows the Theoretical Habit Strengths in Habs of Re-
ceptor-Effector Conjunctions With Unlimited Practice When the Reac-
tion Is Followed by Reinforcement After Varying Amounts of Delay
in Seconds (0- It Is Assumed That With Zero Delay the Reinforcing
Agent Employed Would Yield a Habit Strength (AT) of 80 Habs at the
Limit of Practice, and That Each Additional Second of Delay Reduces
the Limit of Habit Strength by 1/65th.

Amount of
Delay

mmm

Amount of
Delay
(0

Habit-

Strength

Limit

(m')

Amount of
Delay
(0

Habit-

Strength

Limit

(m 1 )

80.00

50.29

31.61

mariiirkmM

49.52

29.26

48.76

27.08

lil'Z&yJM

48.01

25.07

75.20

47.27

23.20

74.04

46.55

21.48

72.91

45.83

19.91

71.79

45.13

18.39

70.68

44.44

100

17.02

69.60

43.75

105

15.76

68.53

43.08

110

14.58

67.48

42.42

115

13.50

66.44

41.77

120

12.49

65.42

41.13

125

11.56

64.42

40.50

130

10.70

63.43

39.87

140

9.17

62.46

39.26

150

7.85

61.50

38.66

160

6.73

60.45

38.07

170

5.76

59.62

37.48

180

4.94

58.71

36.90

190

4.23

57.81

36.34

200

3.62

56.92

35.78

210

3.10

56.04

35.23

220

2.66

55.18

34.69

230

2.28

54.34

34.16

240

1.95

53.50

33.63

270

1.23

52.68

33.12

300

.77

51.87

32.61

330

.49

51.07

32.11

360

.31

of reinforcement and secondary reinforcement are represented in
the goal gradient, which they presumably generate.

In the interest of immediate utilization we accordingly proceed
to the consideration of the more detailed characteristics of the goal

bHb and the delay in reinforcement

*45

gradient. On the basis of the direct experimental approaches out-
lined above, together with certain indirect approaches presently to
be disclosed, we formulate our hypothesis concerning the molar
functional relationship of habit strength to the temporal delay in
reinforcement as follows: (f) The maximum habit strength (m')
attainable with a given amount and quality of reinforcement closely
approximates a negative growth function of the time ( t ) separat-
ing the reaction from the reinforcing state of affairs ; {2) the asymp-

360 330 300 270 240 210 IB0 150 (20 90 60 30 0

DELAY OF REINFORCEMENT IN SECONDS (t )

Fia. 33. Theoretical goal gradient plotted from the values shown in Table
2. This curve is drawn on the assumption that the value of habit strength at
the limit of practice is reduced l/65th for each additional second of delay
in reinforcement. The braces on the vertical scale show the theoretical
difference in habit strength which would be produced by each at the limit of
practice. Note (1) that none of these differences is like any of the others,
and (2) that the middle difference is the greatest of the three.

tote or limit of fall of this gradient is zero; and (3) the more
favorable the condition for the action of secondary reinforcement ,
the slower will be the rate of fall, so that this limit may not be
approximated until after considerable periods of delay, though for
many conditions less favorable for secondary reinforcement it may
be reached in a period of from SO to 60 seconds.

For purposes of precise illustration, the values of such a nega-
tive growth function have been calculated and are reproduced as
Table 2, on the assumption that the conditions of secondary rein-
forcement are such as to bring the gradient practically to zero at
a delay in “primary” reinforcement of about six minutes. The

PRINCIPLES OF BEHAVIOR

146

values appearing in this table are represented graphically by the
curve shown in Figure 33.

With a definite hypothesis as to the quantitative character-
istics of the goal gradient available, the possibility at once arises
of deriving from it, and the other principles of the system, numerous
implications in the form of corollaries. A comparison of these
deductions with relevant experimental evidence then affords a basis
for the acceptance, rejection, or further modification of the hy-
pothesis. With good fortune, this indirect procedure, supplementing
that of the direct experimental attack already considered, may be
expected to lead to the isolation of a sound and comprehensive
scientific principle more quickly than would either approach em-
ployed alone.

ORGANISMS GRADUALLY ACQUIRE A PREFERENCE FOR THE ACT

FOLLOWED BY THE SHORTER DELAY IN REINFORCEMENT

Let it be assumed that hungry albino rats are presented with a
choice of two short passageways; at the end of each is an exactly
similar food reinforcement. In both alleys alike there is, next to
the food box, a delay chamber in which the animal is retained for
a certain period before being permitted access to the food. More
specifically, let it be assumed that the delay in one chamber is
30 seconds and that the delay in the other is 60 seconds. Finally,
during training one or the other alley is always blocked, the order
of the reinforcements on the respective alleys being randomized m
such a way as to keep approximately the same number of rein-
forcements on each at all times. Therefore, at any point in the
training at which it is desired to know the relative strengths of the
two habits thus set up, the entrance to both alleys may be left open,
at which time a competition between the two habit strengths will
occur, the stronger habit of course dominating. By training groups
of comparable animals with different numbers of reinforcements
before the testing, it would be possible to determine relatively how
strong the two habits are at various stages of training, and in this
indirect manner verify empirically the goal gradient hypothesis
formulated above.

Table 2 shows that the theoretical limit ( m ') of habit strength
at a delay of 30 seconds is 50.29 habs, and the limit at 60 seconds
is 31.61 habs. With these asymptotes of the positive growth, or
learning, curves available, and taking our fractional increment ( F )

*47

bHb and the delay in reinforcement

TABLE 3

COLUMNS 2, 3, AND 4 OF THIS TABLE SHOW THE THEORETICAL HABIT STRENGTH

at the Successive Reinforcements by the Same Reinforcing Agent Where
the Fractional Increment (F) Is 1/20, and Where the Delay in Rein-
forcement Is 30% 60% and 90" Respectively. The m' Values Employed
in the Computation of These s Hr Values Were Taken from Table 2.
The V alues in Columns 5 and 6 Represent the Per Cent of Test Trials
in Which the Habit With the Shorter Delay in Reinforcement Would
Be Expected to Dominate on the Assumption That the Range of Oscilla-
tion Has a Standard Deviation of 13 Habs.

Ordinal
Number of
Reinforce-
ment

Strength of
sHr in
Habs at 30'
Delay in
Reinforce-
ment

Strength of
sH r in
Habs at 60'
Delay in
Reinforce-
ment

Strength of
sHr in
Habs at 90'
Delay in
Reinforce-
ment

0.0

2.5

1.6

1.0

4.9

3.1

1.9

7.2

4.5

2.8

9.3

5.9

3.7

11.4

7.2

4.5

13.3

8.4

5.3

16.9

10.6

6.7

20.2

12.7

8.0

23.1

14.5

9.2

27.0

17.0

10.7

30.3

19.1

12.0

33.2

20.8

13.1

36.3

22.8

14.4

39.5

24.8

15.6

41.9

26.4

16.6

43.8

27.5

17.4

Per Cent
of Trials
at Which
the 30'
Habit
Dominates
Over 60'

50.0

52.0
53.9

55.8

57.5

59.1

60.6
63.4

65.8

68.0

70.7

73.0

74.9

76.9

78.8

80.2

81.2

Per Cent
of Trials
at Which
the 30'
Habit
Dominates
Over 90'

50.0

53.3

56.4

59.3

62.0

64.6
66.9

71.1

74.6

77.6

81.2

84.0

86.2

88.4

90.3

91.6

92.5

at 1/20, there may be generated the progressive learning values
shown in columns 2 and 3 of Table 3. For purposes of ready com-
parison, these learning curves are represented graphically in Fig-
ure 34. It will be noted at a glance that the 30-second curve
gradually rises above the 60-second curve, the distance separating

them increasing as the number of reinforcements increase. This
yields our first corollary:

1% ^he shorter the delay in reinforcement , the steeper becomes
the rise of the associated curve of learning.

To be of critical scientific value, a theoretical deduction should
lead to the possibility of comparison with a relevant empirical

148

PRINCIPLES OF BEHAVIOR

observation. Unfortunately, Corollary I as it stands does not
permit an observational check because habit strength, as such,
cannot be observed. However, by combining it with the well-
known principle that the greater the habit strength, the shorter will
be the time of reaction evocation (p. 105), we easily derive a second
corollary which is susceptible of such verification:

II. When a reaction is reinforced after a short delay , the time
required to execute the act will he less than that required to execute

Fig. 34. Parallel theoretical learning curves with the same rate of rise
(F — 1/20) but with different asymptotes as determined by the respective
periods of delay in reinforcement (30 and 60 seconds) as shown in Table 2.
The above curves were plotted from columns 2 and 3 of Table 3.

a comparable act which has had the same number of reinforcements
but in which the delay of the reinforcements has been longer.

Empirical confirmation of the essential soundness of Corollary 2
is furnished by a number of studies, notably one reported by Ander-
son (2). This experiment was set up ( 1 ) in substantially the
manner postulated in the theoretical arrangement just described,
with the exception that the animals were permitted to choose freely
between the two entrances throughout the training process; the
investigator’s labor was thereby greatly economized. Early in the

bHjc AND THE DELAY IN REINFORCEMENT 149

training this procedure would, of course, begin to give a dispro-
portionately large number of reinforcements to the reaction asso-
ciated with the shorter delay. While this doubtless weakened,
relatively, the act involving the longer delay in reinforcement, it
probably did not materially change the outcome so far as the
present set of corollaries is
concerned.

Anderson’s animals had
to cross a platform on the
way to the retention cham-
bers, the distance amount-
ing to seven feet. It hap-
pens that in this study two
pairs of delays are reported
which are alike in their
ratio (1:3), and closely sim-
ilar in their empirical dis-
criminability (83 per cent
and 80 per cent after 40
reinforcements), yet the
lengths of the delays are
very different: 10 and 30
seconds as compared with
120 and 360 seconds, the
lengths of the second pair
of delays being twelve times
those of the first. The mean
runway times for each of
the eight days of training
of the respective groups of

animals are shown in Figure 35, where it may be seen at a glance
that the acts associated with the pair of short delays have a lower
mean reaction time throughout the entire training period, which
thus agrees with the corollary.

At this point we must introduce a factor to be taken up in
detail a little later — that of the spontaneous oscillation or vari-
ability in habit strength (p. 304 ff.). It will be sufficient here to
say only that there is reason to believe that the effective strength
of all habits when functioning as reaction potentials is subject to
continuous uncorrelated interferences, presumably mainly from
processes arising spontaneously within the nervous system, and that

Fig. 35. Parallel curves showing differing
locomotor times for the same distance,
training, and discriminability, but markedly
different delay in reinforcement. (Plotted
from Anderson’s published Table 1. 2 P.
424.)

PRINCIPLES OF BEHAVIOR

1 5 °

the magnitudes of these disturbances are distributed approximately
according to the "normal” law of probability. On the assumption
that the oscillations in habit strength are largely uncorrelated, it
follows that the weaker of two competing habit tendencies will fre-
quently be depressed only slightly when the stronger one chances
to be depressed relatively much, with the result that the weaker
habit will dominate on that occasion. Indeed, this failure of the

•

Fio. 36. Graphic representation of the theoretical progressive increase in
the per cent of choices of the alley involving the 30-second delay in reinforce-
ment over that of the 60-second delay (lower curve) and that of a 30-secon
delay over that of a 90-second delay (upper curve). These have been plot e
from the values appearing in the fifth and sixth columns, respectively, o
Table 3.

stronger habit to dominate will occasionally occur so long as the
oscillation ranges of the two habits continue to overlap. Naturally
the overlapping of the two oscillations will be greater (1) the wider
the range of the oscillation and (2) the more nearly equal the two
habit strengths are at the time. The theoretical per cent of the
test trials at which the act associated with the shorter delay woul
dominate has been calculated on the assumption that the standar
deviation of the factors producing oscillation in habit strengt
remains constant at the equivalent of 13 habs. These values are
shown in the fifth column of Table 3 and are represented graphi-
cally as the lower curve of Figure 36.

a H a AND THE DELAY IN REINFORCEMENT 1 5 1

There it will be seen that, theoretically, at first the two choices
are equally likely, i.e., the entrance to the alley leading to the
chamber associated with the 30-second delay should occur on only
50 per cent of the test trials because the two probability distribu-
tions coincide exactly. However, as practice progresses the choice
associated with the shorter delay gradually attains an advantage

Fia. 37. Graph showing empirical curves of increasing preferences for the
reaction involving the shorter of two delays in reinforcement. (Plotted from
tables published by Anderson, 1.)

which, after 40 pairs of reinforcements, reaches the considerable
amount of 81.2 per cent.

On the basis of the above calculations we generalize and formu-
late the following corollaries:

III. With training , organisms tend to choose that one of a pair
of alternative acts which yields reinforcement with the lesser delay.

IV. The preference for that one of a pair of acts involving the
lesser delay in reinforcement is attained gradually as training in-
creases.

Empirical confirmation of Corollaries III and IV is seen in the
study reported by Anderson (/). This investigator found that in

152

PRINCIPLES OF BEHAVIOR

the course of 40 trials involving delays of 30 and 60 seconds respec-
tively, eight albino rats displayed a gain in the choice of the act
involving the 30-second delay which extended from 47 per cent
(approximately chance) to 76 per cent. The curve of this learning
is shown in the lower graph of Figure 37. Moreover, DeCamp (5),
Yoshioka {17), and Grice (4) have all found substantially the
same relationship to hold where the delay in reinforcement was
incidental to the difference in length of two alternative paths lead-
ing to a reinforcement. It is noteworthy that Grice’s experiment
was set up in such a way as to keep the number of reinforcements
on the two paths more nearly equal than was the case in any of
the other studies so far reported; his results are therefore more
comparable to the conditions presupposed by the above deductions.

THE RATE OF DISCRIMINATION OF ACTS AS A FUNCTION (1) OF
THE DIFFERENCE IN THE DELAYS INVOLVED AND (2) OF
THE ABSOLUTE MAGNITUDE OF THE DELAYS

•

Our next problem concerns the relative rate of preference acqui-
sition of alternative reactions involving differential delays in rein-
forcement as a function of the amount of difference in the two
delays. Let us take, for example, the same theoretical arrangement
assumed above, with the exception that the delays are of 30 and
90 seconds respectively, which gives the organism the relatively
coarse ratio of 1 to 3 instead of 1 to 2, as previously. The theo-
retical course of the acquisition of habit strength with a 30 -second
delay in reinforcement may be seen in the second column of
Table 3, and that of a 90-second delay, in the fourth column. The
per cent of the trials in which the act associated with the 30 -second
delay of reinforcement would be expected to dominate over that
associated with the 90-second delay is given in the sixth column
and is represented graphically in the upper (solid) curve of Fig-
ure 36. There it may be seen at a glance that the larger difference
yields the more rapid acquisition of dominance by the act involving
the lesser delay of reinforcement. Specifically, at the fifteenth
reinforcement the discrimination involving the 30 - 90 -second delay
reached the level of dominance attained at the fortieth reinforce-
ment by the one involving the 30-60-second delay. Generalizing,

we arrive at our fifth and sixth corollaries: .

V. Other things equal , the greater the difference in the delay o/
reinforcement of two competing reactions, the less will be the tram -

bHb and the delay IN REINFORCEMENT 153

ing required to give the act involving the lesser delay a given degree
of dominance.

VI. Other things equal , the coarser the ratio of the delay of
reinforcement of two competing reactions , the less will be the train-
ing required to give the act involving the lesser delay a given degree
of dominance.

Corollaries V and VI also find ready empirical verification in
the investigation reported by Anderson (7). For purposes of easy
comparison, the experimental learning scores of Anderson’s 30-60-
second group of animals (a ratio of 1 to 2) are presented graphi-
cally in Figure 37, in parallel with the results from his 30-90-second
group ( a ratio of 1 to 3). While somewhat irregular, as is to be
expected from the relatively small number of animals employed
in the respective groups, the general relationship shown by the two
theoretical curves of Figure 36 is discernible. The 30-90-second
discrimination reached, between 15 and 20 reinforcements, a degree
of short-delay dominance only attained by the 30-60-second com-
bination between the 35th and 40th reinforcements. Completely
parallel results were obtained by Yoshioka (17) in the discrimina-
tion of pairs of alternate paths to a goal. He reports that a given

TABLE 4

Table Showing the Per Cent of Choices of Acts Associated With the
Shorter of Two Delays to Be Expected on Theoretical Grounds and
the Parallel Empirical Values Reported by Anderson ( 1 , p. 54).

Ratio

Delays of
Reinforcement
Compared

Theoretical Per Cent
Choices of Act
Involving Shorter Delay

Empirical Per Cent
Choices of Act
Involving Shorter Delay

1 : 3

120*':360*'

71.8

1 : 3

60':180'

89.7

87*

1 : 3

30':90 r

92.5

1 : 3

10':30'

80.6

1 : 2

120':240'

69.1

1 : 2

60':120'

81.8

1 : 2

30 r :60 r

81.2

1 : 2

10 ': 20 '

67.9

I : 1.5

120M80 r

64.0

1 : 1.5

60':90'

71.0

1 : 1.5

30':45'

68.9

• • •

1 : 1.5

10 r :15'

59.6

• • •

* At one point in Anderson’s article this value is given as 84 per cent, but in the original table,
as well as in his thesis on file in Yale University, the mean of which has been checked, the value is
87 per cent.

PRINCIPLES OF BEHAVIOR

*54

amount of training on two alternative alleys 210 and 233 inches
in length (a difference of 23 inches and a ratio of about 1 to 1.11)
yielded a preference for the shorter path of 10.80 units, whereas
when the 210-inch alley was paired with a 276-inch alley (a dif-
ference of 66 inches and a ratio of about 1 to 1.3), the same amount
of training yielded a preference for the shorter alley of 18.85 units.

Our next problem concerns the relative ease of discriminating
acts involving a given amount of difference in the delay of rein-
forcement as dependent upon whether the length of the shorter of
each pair of delays is absolutely short or long. For example, what
does our hypothesis imply as to the ease of differentiating acts
associated with delays of 30 and 60 seconds as compared with
those of 60 and 90 seconds when both pairs of delays involve ex-
actly the same absolute difference, 30 seconds? We have already
seen (Table 3 and Figure 36) that theoretically 30 and 60 seconds
yield at 40 reinforcements 81.2 per cent choice of the 30 -second
act. Table 4 shows that at 40 reinforcements the 60-second choice
over the 90-second choice, theoretically, occurs on 71.0 per cent of

the trials. But,

81.2 > 71.0.

Generalizing, we arrive at our seventh corollary:

VII. Equal differences in the delays of reinforcement of two
competing acts lead with equal practice to a lower per cent of
preferential choices of the act associated with the shorter delay
when the two delays are large (i.e., when the ratio is relatively
coarse) than when they are small (i.e., when the ratio is finer).

This corollary also finds complete empirical confirmation. In
the Anderson study already cited, it was found (column 4, Table 3)
that at 40 reinforcements delays of 30 and 60 seconds gave 76 per
cent preference for the 30-second act, whereas delays of 60 and
90 seconds gave a preference of only 66 per cent to the 60 -second
act. Similarly, Grice found that alternative alleys of six feet and
twelve feet, with a six-foot difference, were discriminated m a
mean of 12.4 trials, whereas alleys of 24 feet and 30 feet with
exactly the same difference were discriminated only after a mean
of 28.4 trials, i.e., after about twice as much training.

THE RELATION OF THE DELAY IN REINFORCEMENT

TO WEBER’S LAW

Experimental results such as those just summarized under Corol-
laries VI and VII have led to the view that the discrimination

bHb and the delay IN REINFORCEMENT 1 55

between intervals of delay in reinforcement as found by Anderson
(f) and between alternative paths to reinforcement as found by
Yoshioka (17) indicates the conformity of these learning processes
to Weber's law. Indeed, influenced largely by the views of Yoshi-
oka, the present writer first postulated the gradient of reinforce-
ment as being a logarithmic function of the delay of reinforcement,
which is implicit in the Weber’s law hypothesis; further considera-
tion, which revealed certain mathematical paradoxes arising from
the nature of the logarithmic function (7, p. 273) led to its aban-
donment in favor of the exponential or negative growth function
represented in Table 2 and Figure 33.

This problem brings us to our eighth corollary. Weber’s law,
as applied to the delay in reinforcement, requires that all pairs of
delays of equal ratios, e.g., 1 to 2, with equal amounts of training,
yield equal per cents of preference for the act associated with the
shorter delay. The present (exponential) hypothesis leads to quite
different expectations. Suppose that we have four pairs of acts,
all with delays in the ratio of 1 to 2 as follows:

10" vs. 20"; 30" vs. 60"; 60" vs. 120"; 120" vs. 240".

By Table 2, these pairs of delays at the limit of practice generate
the following habit strengths:

68.53:58.71; 50.29:31.61; 31.61:12.49; 12.49:1.95.

Appropriate computations based on the same assumptions as out-
lined above show that at 40 reinforcements the respective situations
would generate the following pairs of habit strengths:

59.71:51.15; 4381:27.54; 27.54:10.88; 10.88:1.70.

Further calculations show that these pairs of habit strengths cor-
respond to the following per cents of preference for the shorter
delays:

10": 20" =67.9 per cent
30": 60" =81.2 per cent
60": 120" = 81.8 per cent
120": 240" = 69.1 per cent

Comparable computations have been made for a series of delays
which stand in a coarser ratio of 1 to 3 and in a finer ratio of 1 to
1.5. The theoretical outcome of all three sets of delays has been
drawn up systematically in Table 4.

*5 6

PRINCIPLES OF BEHAVIOR

An examination of the third column of this table shows that
according to the present set of hypotheses, it is not to be expected
that all pairs of acts whose delays of reinforcement stand in the
same ratio will be equally discriminable. On the contrary, it is
evident that while the several entries under a given ratio show some
resemblance, they also show marked differences. Moreover, these
differences manifest a characteristic pattern. First, there is a cen-
tral point of maximum discriminability from which the ease of dis-
crimination decreases as the absolute magnitude of the delays either
increases or decreases; thus in the ratio of 1:3 the combination of
30-90 seconds gives a maximum of 92.5 per cent, whereas that of
60-180 seconds, a pair of larger values, gives 89.7 per cent, and that
of 10-30 seconds, a pair of smaller values, gives 80.6 per cent.
Second, as the ratio of the delays grows finer, the point of maxi-
mum ease of discrimination shifts in the direction of the longer
delays; thus the maximum ease of discrimination falls in the 1:3
ratio on the combination of 30-90 seconds, whereas at the 1:2 ratio
it falls on the larger values of 60-120 seconds.

These a priori expectations find striking empirical confirmation
in the experimental results reported by Anderson (1). The rele-
vant empirical values have been arranged in column 4 of Table 4,
in parallel with the theoretical values. There it may be seen that

(1) the discriminability of acts associated with different delays in
reinforcement is in fact not equal, i.e., Weber’s law does not hold;

(2) the combination 30-90 seconds has the maximum ease of dis-
criminability of the 1:3 ratio exactly as calculated, from which the
ease of discrimination falls off in both directions; (3) the combina-
tion 60-120 seconds has the maximum ease of discriminability of
the 1:2 ratio, a shift in the direction of the larger values, from
which the ease of discrimination diminishes in both directions,
again exactly as demanded by the theory.

A similar outcome in favor of the exponential and opposed to
the logarithmic relationship was obtained by Grice (4) in the ease
of discriminating different ratios of alternate paths to food rein-
forcement. While Grice’s experiment was not designed in such a
way as to bring out the exponential relationship as dramatically as
does Anderson’s study, this characteristic in his results was effec-
tively demonstrated by means of a process of curve fitting.

Generalizing from the preceding observations, we may formulate

our eighth corollary as follows:

bHm and the delay in reinforcement

>57

VIII. The ease of discrimination of acts associated with dif-
ferent delays of reinforcement has for a given ratio of delay lengths
a maximum occurring at a central point of absolute lengths , from
which point it diminishes progressively with both increase and de-
crease in the absolute magnitude of the delays, the point of maxi-
mum ease of discrimination shifting in the direction of the longer
delays as the ratio of the two delays grows finer.

Extrapolating from the same general considerations, we may
formulate a ninth corollary:

IX. With the ratio between delays of reinforcement constant ,
discrimination becomes impossible when the periods of delay in-
volved become sufficiently great or sufficiently small.

SUMMARY

The favorite method employed by experimentalists in determin-
ing the functional relationship of the rate of learning to the delay
in reinforcement has been to present an organism with the pos-
sibility of performing two alternative acts one of which receives
reinforcement after a shorter delay than the other. The organism
is then permitted gradually to develop a preference for the act
involving the shorter delay by a process of trial and error, the
number of trials required to reach a given preference criterion being
taken as an indication of the ease of the “discrimination.” In the
purer forms of these experiments the acts are strictly comparable,
e.g., merely turning to the right or the left and walking a few inches
or feet. The alternative acts of some experiments, however, require
the organism to traverse different lengths of paths to reach the
point of reinforcement, in which case not only is there presumably
involved the delay required to traverse a given path, but there is
introduced a third, though closely parallel factor, the amount of
work or energy expenditure required in traversing the different
distances (see p. 280 ff.).

The early experimental work on the relation of the rate of
learning to the extent of the delay in reinforcement yielded nega-
tive results. This was apparently because in the procedures em-
ployed, the food box served as the retention chamber, which intro-
duced in a gross manner the factor of secondary reinforcement.
When a separate compartment adjoining the food box was used
as a retention chamber, the spread of secondary reinforcing ten-
dencies was sufficiently gradual to show in one study, that of

158 PRINCIPLES OF BEHAVIOR

Wolfe, a negatively accelerated falling gradient extending to delays
of reinforcement 20 minutes in duration. Perm, on the other hand,
in an experiment set up in such a way as to exclude secondary
reinforcement as completely as possible, found a negatively accel-
erated gradient of rate of learning which fell to zero at a delay of
only 30 seconds or so.

The most plausible interpretation of these superficially conflict-
ing results seems to be (1) that the basic or primary gradient of
reinf or cement is only about 30 seconds in duration, and (2) that
under ordinary learning conditions secondary reinforcement com-
bines with the gradient of reinforcement to produce a derived phe-
nomenon which may be called the goal gradient.

Partly on the basis of direct experimental evidence, but mainly
on the grounds of indirect evidence, ft is concluded that the most
plausible hypothesis concerning the quantitative characteristics of
the goal gradient is: In situations where both (1) the primary
gradient of reinforcement and (2) the principle of secondary rein-
forcement are operative in a progressive manner, the goal gradient
is an exponential or negative growth function , and the greater the
influence of secondary reinforcement , the less steep will he the slope
of the gradient.

From this postulate or hypothesis, together with other principles
of the system, a number of corollaries follow. The first of these is
that, other things equal, the shorter the delay in the reinforcement
of a given act, the steeper will be the curve of learning of that ac
and so, at any given number of reinforcements, the shorter will be
the time required to execute the act; this is in agreement wit

experimental findings.

In the typical trial-and-error situation involving alternative
acts with different delays in the reinforcement of each, it follows
from the goal gradient hypothesis as here formulated that the act
associated with the shorter delay of reinforcement will acquire
habit strength at a faster rate than will the alternative act. How-
ever, because of the principle of oscillation, the stronger of t e
two habits would be expected to attain dominance, not at once bu
only gradually, and often imperfectly, even after a very large nuna-
ber of trials. This a priori expectation is confirmed by experimen •

On the same principle it is to be expected that, other things
p Q ual the larger the difference in the delays associated with ®

competing acts, the fewer will be the trials required ^
given per cent of dominance. Similarly, for a given differenc

bHm and the delay in reinforcement

>59

the delays associated with the respective acts, the longer the actual
delays involved, the greater will be the number of trials required
to produce a given per cent of dominance. Both these corollaries
also are confirmed by experiment.

Perhaps the most critical test of the hypothesis concerns the
ease of learning to discriminate comparable acts associated with
delays of equal relative, but of different absolute, duration. The
exponential goal gradient hypothesis as here formulated implies
that the discriminability of acts associated with delays of reinforce-
ment standing in the same ratio to each other will be maximal at
a central region of absolute delays, but at other delays, whether
greater or less, the ease of discrimination declines progressively.
Moreover, it follows from these same principles that the point of
maximum ease of discrimination of a coarse ratio, such as 1 to 3,
will appear at smaller absolute values than will be the case with
a fine ratio, such as 1 to 2. All of these intricate and detailed im-
plications of the exponential goal gradient hypothesis are in re-
markable agreement with the experimental observations at present
available. This fact goes far to indicate that the goal gradient is
a negative growth, or exponential, function. Incidentally this fur-
nishes an excellent example of the manner in which indirect pro-
cedures may sometimes yield the characteristics of a function
which has proved refractory to direct attack.

NOTES
Historical Note

The first mention of the gradient-of-reinforcement principle which we have
been able to find was by Thorndike, in 1913. At that time he wrote: “Such
intimacy, or closeness of connection between the satisfying state of affairs and
the bond it affects, may be due to close temporal sequence. . . Other things
being equal, the same degree of satisfyingness will act more strongly on a bond
made two seconds previously than on one made two minutes previously. . .”
(11, pp. 172-173). Thirteen years later, Margaret Washburn remarked in the
third edition of The Animal Mind : “The facts show that it [the drive] will set
most strongly in readiness those movements which most immediately preceded
its resolution on a previous occasion. This ‘gradient' of excitation from move-
ments just before the final ‘success,’ step by step to those at the beginning of the
series, may also be explained by the ordinary associative laws. The movements
nearest the end of the series have a greater readiness due to recency of perform-
ance.” (14, P- 335).

In 1932 the same general concept was put forward by the 'present writer in
an attempt at a deductive explanation of numerous molar phenomena of animal

i6o

PRINCIPLES OF BEHAVIOR

maze learning, such as the preference of the shorter path to a goal, the fact of
blind alley elimination, the backward order of such eliminations, and so on:
“The mechanism which in the present paper will be mainly depended upon as an
explanatory and integrating principle is that the goal reaction gets conditioned
the most strongly to the stimuli preceding it, and the other reactions of the be-
havior sequence get conditioned to their stimuli progressively weaker as they are
more remote (in time or space) from the goal reaction. This principle is clearly
that of a gradient, and the gradient is evidently somehow related to the goal
We shall accordingly call it the goal gradient hypothesis .” ( 6 , pp. 25-26.)

In order to perform these deductions it was necessary to postulate the mathe-
matical characteristics of the gradient of habit strength which both Thorndike
and Washburn had specified in a general way. Largely because of the results of
Yoshi oka's investigation (17), habit strength was postulated as being a function
of the logarithm of the amount of time (0 or space separating the receptor-
effector conjunction from the reinforcing state of affairs, i.e.,

sH r = a — b log t.

In 1938, however, the logarithmic gradient was rejected on the grounds (1) that
when t — 0, sHr becomes infinite, and (2) that with large values of t, rHr becomes
negative, both of which seem a priori improbable (7, p. 273). Accordingly the
logarithmic equation was replaced by an exponential equation. In terms of our
present notation this is:

m’ = M'e -* ; (13)

where M’ is the learning asymptote under a given reinforcing agent, t has the
same significance as in the logarithmic equation, and m' is the learning asymptote
with a given delay in reinforcement and reinforcing agent. The exponential
equation, because of the excellent agreement of its implications with a consider-
able range of empirical findings, is regarded as the closest approximation to the
goal gradient function at present available.

The Wolfe Data

Fortunately, Wolfe (16) published the per cent of correct runs for each of
his eight groups of rats at each of the ten days of the training process. An exami-
nation of these data suggested that the scores of days 7, 8, 9, and 10 are the most
stable and significant of the series for our present purposes. Accordingly these
probability-of-success scores for the four days were pooled for each group o
animals. # ,

The next step was to convert these probability-of-success scores into units of
amount on some linear scale. This was done on the assumption (see Chapter
XIII) that learning manifests itself in this experiment only by overriding an
oscillatory tendency varying from moment to moment, the net result of whic »
on the average, is to give about equal numbers of choices to the right and the le
paths. This chance oscillation factor is assumed further to distribute itse
approximately according to the Gaussian or “normal” law of chance. ^ Con-
venient tables of this function have been derived by mathematicians, where y
any per cent of probability may be at once converted into amounts in presuma y
Unpor units The unit in such cases is usually the standard deviation w ° .

of the chance variability involved. A simplified table of tins kind

bHb and the delay IN REINFORCEMENT l6l

is shown as Table 9 on page 311. However, a more elaborate table was used
with the present data. The use of the table may be illustrated as follows: the
pooled scores of days 7, 8, 9, and 10 of the one-minute group show 64.9 per cent
of correct reaction. Referring to our table we find that this corresponds to a shift
.40 <r from the pure chance (50-50) distribution of choices in the correct direction.
In this way were obtained the following <r-values which may be regarded as
indices of habit strength :

Amount of delay

Index of habit strength at
trials 7, 8, 9, and 10

0 r

1.21 a

.98 <r

30'

.49 <r

.40 a

2.5'

.42 <r

.28 <r

10'

.30 a

20'

.18(7

Figure 30 is plotted from these values. The negative growth function fitted to
the first four of the values is :

m' = .375 + (1.21 - .375) e - - 068 *.

The smooth curved line of Figure 31 is plotted from this equation.

Some indirect confirmation of the hypothesis that more than one factor is
operating to produce Wolfe’s empirical gradient is furnished by the fact that it
was found possible to get a rather satisfactory fit to the above set of values by
assuming that they represent the simple summation of two growth functions, a
major one corresponding roughly to the equation given above, and a minor one
representing a rather slow learning process with a gentler slope. The complete
equation, including both of the supposed components, is :

m' = .175 + .225 e- oon t + 810 e -.065

The Characteristics of Perm's Empirical Gradient

The curve which has been fitted to Perm's empirical gradient of reinforcement
is a diminishing exponential function which has as its asymptote, not a horizontal
straight line as have ordinary exponential functions, but a straight line which slopes
downward in such a way as to cross the horizontal axis. This equation, shown
graphically in Figure 32, is :

N = 40

tan 2 A b H b = 1.6 X 10“-« < — .043 t + 1.45,

N « 60

N = 40

in which tan 2 A sHr represents the tangent of the curve of learning at the

N = 60

point of 50 per cent choice of the act which removes the manipulative bar. This
it will be noticed, is by no means the same as habit strength ( s //*). Moreover]
there is reason to believe that appreciable amounts of extinction effects con-

\6i

PRINCIPLES OF BEHAVIOR

tribute to this function which may account for the fact that extrapolation where
t > 34 yields negative values.

Does the Delay-of-Reinforcement Gradient Extend Forward as Well as

Backward from the Point of Reinforcement?

Thorndike and his associates have reported evidence which suggests that
the gradient of reinforcement may extend also in the forward direction; i.e.,
that learning may be reinforced when the reinforcing state of affairs precedes
the S — R conjunction as well as when it follows it. Recently Jenkins ( 8 ) has
confirmed these results in a convincing manner, using albino rats as subjects
in a maze situation. These studies indicate that the maximum of the forward
gradient is considerably lower than that of the backward gradient. The adaptive
significance of this second gradient has not as yet been very carefully studied.
However, since the reduction in a need necessarily follows rather than precedes
the act which brings this about, it would seem that the forward gradient could
hardly play much role in selective learning.

Yoshioka’s Study of Path-Length Discrimination in Rats

A number of years ago, Yoshioka (17) investigated experimentally the relative
difficulty of setting up a consistent preference for the shorter of two paths to a
food goal. In general he found that the ease of setting up the short-path prefer-
ence was approximately the same for two mazes, one of which was twice as large
as the other, for five different ratios of long to short path in each maze.

But how can Yoshioka’s results be reconciled with the gradient taken as an
exponential function of the form,

m' = M’eri*?

We have already shown above (p. 156) that if two pairs of delays in reinforcement,
one pair twice as long as the other, are chosen at certain points at each side of a
central point of maximum discriminability, one pair will be approximately as
easy to learn as the other. It is evident that there may be found a very lari5®
number of such pairs of alternative delays. It is possible that Yoshioka chanced
to select a series of pairs of alternate pathways involving just such delays m
reinforcement.

Equations From Which the Various Theoretical Curves and Tables Have

Been Derived

The exponential or negative growth function of the goal gradient from which
Table 2 and Figure 33 were derived is,

m' = M' X lO" 00672 *, ( 14 )

where Af ' is the maximum strength attainable with an unlimited number of rein-
forcements with the reinforcing agent employed. . .

The equation from which Table 3 and the curves of Figure 34 were derived is,

g H B - m' — to' X 10 - 02226 y ,

where m' is as defined by the preceding equation. . ... the

The method of deriving the per cent of choices of the act associated
shorter delay of reinforcement is first to calculate the strength of the respe

sHm AND THE DELAY IN REINFORCEMENT 163

habits {Ht and Hi) by means of equations 1 and 2;
in equation 15:

Hr - Hi

V*1 4- ol '

then substitute these values

where <r x is the standard deviation of the oscillation of H u and a is the standard
deviation of the oscillation of H t . Substitute the values of H lf H t , <j u and a,
in this equation, solve for x, and look up the value of p, probability of occurrence
or per cent of dominance of the stronger habit, in a table of the probability integral.

Example: Let us take the habit strengths yielded by delays of 30 and 60
seconds at 40 reinforcements (Table 3, columns 2 and 3). These turn out to be
43.8 habs and 27.5 habs respectively. Also it will be recalled that the oscillation
range of both habits was assumed to have a standard deviation of 13 habs. Sub-
stituting these values in equation 3, we have,

43.8 - 27.5
X ~ Vl3 2 + 13 2

_ 163
~ V338

x = .886.

Looking on page 76 of Kelley’s Statistical Tables at x = .886, it is found that
corresponding to this>alue is a p value of .812 ; *.e., the stronger of the two habits
will dominate in the long run 81.2 per cent of the test trials, exactly as is shown
at the bottom of column 5.

REFERENCES

1. Anderson, A. C. Time discrimination in the white rat. J. Comp.

Psychol., 1932, IS, 27-55.

2. Anderson, A. C. Runway time and the goal gradient. J. Exper. Psychol.,

1933, 16, 423-428.

3. DeCamp, J. E. Relative distance as a factor in the white rat’s selection

of a path. Psychobiology, 1920, 2, 245-253.

4. Grice, G. R. An experimental study of the gradient of reinforcement in

maze learning. J. Exper. Psychol., 1942, SO, 475-489.

5. Hamilton, E. L. The effect of delayed incentive on the hunger drive in

the white rat. Genet. Psychol. Monogr., 1929, 5, 131-207.

6. Hull, C. L. The goal gradient hypothesis and maze learning. Psychol.

Rev., 1932, 39, 25-43.

7. Hull, C. L. The goal gradient hypothesis applied to some ‘field-force’

problems in the behavior of young children. Psychol. Rev., 1938 45.
271-299.

8. Jenkins, W. O. Studies in the spread of effect. PhD. thesis, Yale Uni-

versity, 1942.

9. Miller, N. E., and Miles, W. R. Effect of caffeine on the running speed

of hungry, satiated, and frustrated rats. J. Comp. Psychol., 1935, 20,
397-412.

10. Perin, C. T. The effect of delayed reinforcement upon the differentiation
of bar responses in white rats. J. Exper. Psychol , 1943, 52, 95-109.

11* Thorndike, E. L. Educational psychology, Vol. I. The original nature
of man. New York: Teachers College, Columbia Univ., 1913.

1 64 PRINCIPLES OF BEHAVIOR

12. Thorndike, E. L. The fundamentals of learning. New York: Teachers

College, Columbia Univ., 1932.

13. Warden, C. J., and Haas, E. L. The effect of short intervals of delay in

feeding upon speed of maze learning. J. Comp. Psychol., 1927, 7,
107-116.

14. Washburn, M. F. The animal mind (third edition). New York:

Macmillan, 1936.

15. Watson, J. B. The effect of delayed feeding upon learning. Psycho-

biology, 1917, 1, 51-60.

16. Wolfe, J. B. The effect of delayed reward upon learning in the white

rat. J. Comp. Psychol., 1934, 17, 1-21.

17. Yoshioka, J. G. Weber’s law in the discrimination of maze distance by

the white rat. Univ. Calif. Pub. in Psychol., 1929, 4, 155-184.

CHAPTER XI

Habit Strength as a Function of the Temporal Relation
of the Conditioned Stimulus to the Reaction

Each of the last three chapters has been concerned with the
quantitative aspects of one of the antecedent conditions of rein-
forcement which determine habit strength. In this, the fourth chap-
ter on this general subject, we shall consider the functional de-
pendence of habit strength upon a pair of closely related antecedent
conditions. These have already been laid down (p. 71) as the
necessary qualitative conditions for learning, namely, that there
must be a temporal contiguity between an effector activity and (1)
an afferent impulse or (2) the perseverative trace of such an im-
pulse. We shall begin with the consideration of the former.

HABIT STRENGTH AS A FUNCTION OF THE DURATION OF
THE CONDITIONED STIMULUS AT THE TIME

OF REACTION OCCURRENCE

The question of the rate of habit formation as a function of
the time the conditioned stimulus ( S ) has been acting when the
reaction ( R ) occurs has been submitted to systematic study by
Kappauf and Schlosberg (5). These investigators delivered the
unconditioned stimulus, in the form of a 1/3-second electric shock,
to the right front leg of each of a series of albino rats. The con-
ditioned stimulus was a loud buzzer which temporally overlapped
the shock. With different groups of animals the buzzer began 1 /3,
2/3, 1, 2, 4, and 7 seconds before the shock, both stimuli terminat-
ing at the same time. Accordingly the habits thus set up would,
in Pavlov’s terminology ( 6 , p. 88), be called “delayed” conditioned
reflexes, as contrasted with “trace” conditioned reflexes (ff, p. 40)
in which the action of the conditioned stimulus terminates before
the onset of the unconditioned stimulus.

Kappauf and Schlosberg recorded and measured several differ-
ent responses to shock which it was thought might become condi-
tioned. Possibly because of the small number of animals employed
in each group and the fact that each group was subdivided by
differential treatment, the various reactions yielded somewhat dis-

165

1 66

PRINCIPLES OF BEHAVIOR

cordant results. The single response which gave the most consistent
conditioned reactions was a sharp inspiration or gasp. By pooling
measurements taken from published graphs representing the scores
of the two animals making up each of the six delay groups, there
has been obtained an indication of what purports to be the func-
tional relationship which we are seeking. The values so secured
are represented by the circles of Figure 38.

0 1 2 3 4 5 6 '

NUMBER OF SECONDS § c PRECEDES $ u

Fio. 38. Graphic representation of the per cent of conditioned stimuli
evoking antedating reactions in a delayed conditioned reflex, as a function °
the time interval from the beginning of the conditioned stimulus to the onset
of the unconditioned stimulus. (Plotted from pooled measures of two 6™phs
showing per cent of conditioned gasping reactions in rats, published t>y
Kappauf and Schlosberg 5 .)

It is at once evident from an inspection of the arrangement of
these circles that long-delayed conditioned stimuli are much less
effective in acquiring receptor-effector connections under reinforce-
ment conditions than are those of relatively short delay. More-
over, beginning at the delay of maximum efficiency there appears to
be a progressive falling off in the per cent of stimulations evoking
conditioned reactions, the decline taking place approximately ac-
cording to a simple negative growth or decay function of the de ay
beyond the optimum. Such a function was fitted to the va ues

b H k AND STIMULUS-RESPONSE ASYNCHRONISM

167

corresponding to the circles, and is represented by the curve drawn
among them. The asymptote or ultimate level of fall turns out
to be 30.4 per cent. This suggests that generalized response sensiti-
zation due to shock was occurring in this experiment, or that an
auditory stimulus will not wholly lose its capacity for becoming
conditioned to a reaction however much it may be prolonged, or
both.

Turning to the left-hand side of Figure 38, there is found a
definite suggestion that learning is facilitated by having the uncon-
ditioned stimulus occur a fraction of a second later than the con-
ditioned stimulus. The Kappauf-Schlosberg experimental technique
makes difficult the distinction between conditioned and uncondi-
tioned reactions in this region; therefore the discussion of this point
will be taken up in connection with an investigation presently to
be considered, which does not suffer from such a handicap.

From the preceding analysis, then, we conclude that within the
limits of the Kappauf-Schlosberg investigation:

1. The maximum efficiency of conditioning occurs when the onset of
the unconditioned stimulus follows that of the conditioned stimulus by a
fraction of a second.

2. As the delay in the onset of the unconditioned stimulus increases
beyond that yielding the maximum learning efficiency, the rate of habit-
strength acquisition decreases progressively according to a simple decay
function of the amount of this additional delay.

3. The value of this function at the limit of its fall is about a third
of that at its highest point. This asymptotic value presumably ap-
proximates the status in learning situations of static, i.e., non-changing,
stimulus elements.

A NEUROLOGICAL HYPOTHESIS AND SOME COROLLARIES

The results shown in Figure 38 present such a striking parallel
to Figures 4 and 5, particularly to Figure 4, that it has been con-
sidered worth while to give brief consideration to some implications
of a related neurological hypothesis which has been suggested by
Kappauf and Schlosberg (5, p. 39). The relevant neurological
phenomena may be summarized briefly as follows:

1. Receptor discharge impulses begin an appreciable interval after the
impact of the stimulus energy on the receptor (J, p. 116; 2 ).

2. As the energy impact on the receptor becomes weaker, the discharge
latency becomes longer (see Figure 3).

i68

PRINCIPLES OF BEHAVIOR

3. Following the period of receptor-discharge latency there is a period
of relatively rapid recruitment in the frequency of receptor discharge
impulses, which usually reach their maximum within a second (1, p. 116).

4. The amount of stimulus energy ultimately applied remaining con-
stant, the faster its rate of application, the more rapid will be the rate
of recruitment and the greater the maximum frequency of receptor-dis-
charge impulse ( 1 , p. 75).

5. If the rate of stimulus impact is relatively abrupt and constant,
the greater the stimulus energy applied, the greater will be the maximum
frequency of receptor impulse discharge (1, p. 116).

6. Following the attainment of the maximum frequency of receptor
discharge impulses, the stimulus meanwhile continuing to act unchanged,
there ensues a progressive decline in frequency approximately according
to a simple decay function. In some receptors, such as touch and those
associated with hairs, the frequency quickly falls to zero; in others, such
as pressure and those associated with muscle spindle, the frequency be-
comes constant at a level well above zero ( 1 , p. 79).

7. In case the impact of the stimulus energy on the receptor organ
ceases before the point of maximum receptor-discharge frequency is
reached, there is usually a brief after-discharge which apparently may be
prolonged under certain circumstances by self-propagating central pro-
cesses (5). This latter perseverative activity presumably declines as a
simple negative growth function of the time since stimulus termination,
the asymptote of this decline being zero (Figure 5, p. 43).

We now add to this summary of empirical findings a formalized
statement of Kappauf and Schlosberg’s hypothesis 1 : Other things
equal, the increment to the strength of a receptor-effector connec-
tion (A ,H r ) resulting from a reinforcement is an increasing func-
tion of the frequency of the associated receptor discharge , or the
intensity of the resulting afferent impulse.

Our immediate concern here is with the implications of the
Kappauf-Schlosberg hypothesis and certain items of the receptor
impulse summary presented above, namely, items 3, 4, 5, 6, and 7.

I. It follows from the above hypothesis and empirical item 3
that in a reinforcement situation there is a temporal relationship
of the conditioned to the unconditioned stimulus such that as the
onset of the unconditioned stimulus is progressively delayed, the
rate of learning will increase.

II. It follows from the hypothesis and empirical item 4, other
factors remaining constant, that in a reinforcement situation the

i Kappauf and Schlosberg are in no way to be held accountable for the
shortcomings of this formulation; the present author assumes entire respo
sibility.

sHb AND STIMULUS-RESPONSE ASYNCHRONISM 109

slower the rate of application of the conditioned stimulus energy ,
the slower will be the rate of habit-strength acquisition.

III. It follows from the hypothesis and empirical item 5 that
in a reinforcement situation, temporal relationships of conditioned
and unconditioned stimuli remaining optimal, within moderate
ranges the greater the conditioned stimulus energy applied, the
more rapid will be the rate of habit-strength acquisition.

IV. It follows from the hypothesis and empirical item 6 that
in a reinforcement situation as the time of the onset of the uncon-
ditioned stimulus is further retarded beyond the point of optimal
timing, the rate of learning will decline, but at a rate slower than
the rate of rise during the recruitment period, the course of the
decline following a negative growth function of the amount of
delay, with asymptote appreciably above zero in the case of cer-
tain receptors.

V. It follows from the hypothesis and empirical item 7 that
in a reinforcement situation as the time of the onset of the uncon-
ditioned stimulus is retarded beyond the optimal amount, the action
of the conditioned stimulus having ceased before the maximum rate
of receptor discharge impulse is reached , trace conditioned reactions
will be generated. In such cases the rate of learning will decline
according to a simple negative growth function of the amount of
delay, with its asymptote at zero.

We may now briefly compare the above deductions with the
facts of habit-strength acquisition. Corollary III is in agreement
with the empirical observations as reported by Pavlov ( 6 ) . As yet
no evidence has been found concerning Corollary II. Corollaries
I and IV are in good qualitative agreement with the empirical
results of Kappauf and Schlosberg. At the best the above deduc-
tions may constitute the beginning of a passage of the molar theory
of behavior over into the ultimate molecular behavior theory based
on neurophysiology; at the worst they may be no more than a
harmless failure in the long trial-and-error history which must
precede the evolution of a true molecular theory of learned be-
havior.

HABIT STRENGTH AS A FUNCTION OF THE DURATION OF THE
STIMULUS TRACE AT THE TIME OF ACTION OCCURRENCE

The fifth corollary derived from the Kappauf and Schlosberg
hypothesis, since it concerns the rate of learning as a function of

1 7° PRINCIPLES OF BEHAVIOR

the age of the stimulus trace which is contiguous with the reaction
to be associated in the learning process, leads directly to the sec-
ond factor whose relation to rate of habit-strength acquisition is
to be considered in the present chapter. A study bearing directly
on this point has been published by Helen Morrill Wolfle.

In one of her two experiments in this field ( 12 , 13 ), Mrs. Wolfle
employed 90 human subjects in groups of ten, each group devoted
to the determination of the effectiveness of trace conditioned-reflex
learning with a particular temporal relationship of the conditioned

Fia. 39. Graphic representation of habit strength as a function of the
relation of the conditioned to the unconditioned stimulus in a
short trace conditioned reaction. Both curves are simple negative growth

functions whose asymptotes are 6.5. (From data published by Helen Morrill
Wolfle, 12 .)

to the unconditioned stimulus. The former was a single sharp
click, the latter, an electric shock to the hand. The reactions
recorded were hand-withdrawal movements associated with shock
avoidance. At irregular intervals throughout the reinforcement
process the conditioned stimulus was presented without the accom-
panying shock. The measure of learning was the per cent of hand
movements following these presentations.

The results of this investigation are shown by the series of
circles in Figure 39. A glance at these circles suggests a confirma-
tion of Corollary I, the maximum efficiency appearing when the

bHz AND STIMULUS-RESPONSE ASYNCHRONISM 1 7 1

shock followed the click by a fraction of a second. This we shall
call the point of optimal stimulus asynchronism. In this the trace
conditioned reaction of Mrs. Wolfle agrees substantially with the
delayed conditioned reaction of Kappauf and Schlosberg.

Corollary V is concerned with the hollow circles, which repre-
sent learning efficiency when the shock, and so the reaction and
its reinforcement, followed the click by a half second or more. An
examination of this portion of Figure 39 reveals that as the onset
of the shock is retarded beyond the point of optimal stimulus
asynchronism the habit strength diminishes consistently, the rate
of diminution decreasing as zero efficiency is approached, quite in
agreement with the corollary. The progressive decline in habit
strength on this side of the point of optimal learning efficiency
we shall call the posterior stimulus asynchronism gradient.

In order to test the corollary in more detail a simple negative
growth function was fitted to these latter values; this is repre-
sented by the broken line passing among the hollow circles. Not-
withstanding the usual deviations, the circles fall fairly close to the
line, which indicates a reasonably good fit. The limit of fall of
this function turns out to be 6.5, a value appreciably above zero.
In this respect empirical results appear to disagree with Corollary
V; however, the failure of this stimulus-asynchronism gradient to
fall to zero may be due to sensitization effects, i.e., the effects of
the shock alone quite apart from its association with the condi-
tioned stimulus (4, p. 431).

THE PROBLEM OF BACKWARD CONDITIONING

“Backward” conditioning is said to take place where the stim-
ulus originally evoking the reaction, and usually the reaction itself,
occur before the impact of the conditioned stimulus. This order of
occurrence during the reinforcement process is called backward
because it is the reverse of the order of occurrence when the
acquired receptor-effector connection functions; in the latter situa-
tion, of course, the stimulus must precede the reaction it evokes.

At one time Pavlov ( 6 , p. 27) regarded backward conditioning
as impossible; later he revised this opinion (7, p. 381), holding that
while backward conditioning is possible, the results obtainable by
this procedure are very weak and unstable. Upon the whole the
latter view has been substantiated by more recent studies ( 9 ),
including two experiments by Mrs. Wolfle (12, 13).

* 7 *

PRINCIPLES OF BEHAVIOR

The two extreme left-hand circles of Figures 39 represent the
results attained by backward conditioning. In this connection it
may be noted that these two circles occupy a position on a smooth
and consistent gradient descending from a point near that of maxi-
mum learning efficiency. We shall call this the anterior stimulus -
asynchronism gradient. This continuity suggests that so-called
“backward” conditioning may be physiologically no more than a
special and extreme case of a gradient antedating not the onset of
the conditioned stimulus but, rather, the optimal phase of stimulus
asynchronism. In order to test the hypothesis a simple negative
growth function was fitted to the four values represented by the
four solid circles. This is shown graphically by the smooth curve
drawn through these circles; the fit may be seen to be nearly per-
fect. Whether or not it be a coincidence, the asymptote of this
gradient is exactly the same as that of the values at the right,
namely, 6.5 per cent. As in the case of the posterior stimulus
asynchronism gradient, the small positive asymptotic value may
very well be due to sensitization effects U, p. 831).

As a final word concerning the data represented in Figure 39
it may be added that the two gradients were extrapolated upward
to where they intersect, on the assumption that the point of inter-
section would indicate indirectly somewhat more precisely the
optimal temporal relationship of the conditioned to the uncondi-
tioned stimulus. The outcome of this manoeuvre is shown graphi-
cally in Figure 39. It suggests that under the conditions of Mrs.
Wolfle s experiment the conditioned stimulus should antedate the
onset of the unconditioned stimulus by .44 second if maximum
learning efficiency is to be attained. It also suggests that had a
group of subjects been conditioned at this interval the learning
yield would have been 42.4 per cent of reaction evocations, an
appreciable advantage over the value obtained at a delay of .5
second.

The results of the above analysis are in agreement with a con-
siderable body of experimental evidence which indicates that for
optimal learning to take place the conditioned stimulus should
precede the onset of the unconditioned stimulus by something less
than a half second. A certain amount of speculation has arisen
regarding the cause of this now well-established fact. One of the
most ingenious of these hypotheses has been put forward by Guth-
rie (5), who has suggested that in the learning situation the stim-

sH a AND STIMULUS-RESPONSE ASYNCHRONISM

*73

ulus actually conditioned to the reaction is the proprioceptive stimu-
lation arising from reactions, usually implicit, evoked by the
conditioned stimulus at the outset of the learning. This obviously
ad hoc hypothesis accounts for the optimal time relationship of
the onset of conditioned and unconditioned stimulation, though it
fails to explain why proprioceptive stimuli should have a monopoly
in the acquisition of receptor-effector connections over discharges
from other important receptors such as the ear and the eye. It
seems likely that the solution of the problem must await the future
developments of neurophysiology; fortunately this is not a neces-
sary prerequisite to the development of a molar system of behavior
theory.

“short” versus “long” trace conditioned reactions

Intimately related to the gradient of habit strength as a func-
tion of the age of the trace at the time of reaction occurrence is
Pavlov’s distinction between short and long trace conditioned re-
flexes. In this connection Pavlov remarks ( 6 , p. 40) :

Trace reflexes may be of different character, depending on the length
of pause between the termination of the conditioned stimulus and the
appearance of the unconditioned stimulus. When the pause is short,
being a matter of only a few seconds, then the trace left by the condi-
tioned stimulus is still fresh, and the reflex is what we may term a short-
trace reflex. On the other hand, if a considerable interval, one minute or
more, is allowed to elapse between the termination of the conditioned and
the beginning of the unconditioned stimulus we have a long-trace reflex.
. . . every stimulus must leave a trace on the nervous system for a
greater or less time — a fact which has long been recognized in physiology
under the name of after effect.

The referent of the term trace in the last sentence quoted above
is evidently the same as that of the expression perseverative stim-
ulus trace in the present work. We thus arrive at the identification
of Mrs. Wolfle’s right-hand gradient (Figure 39) with Pavlov’s
“short” trace conditioned reflex.

Pavlov is less specific about the basis for his “long” trace
conditioned reflexes, though his experimental examples and asso-
ciated remarks make the picture fairly clear ( 6 , pp. 41-42) :

This may be illustrated by the following detailed experiment of Dr.
Feokritova: A dog is placed in a stand and given food regularly every
thirtieth minute. In the control experiments any-one feeding after the

*74

PRINCIPLES OF BEHAVIOR

first few is omitted, and it is found that despite the omission a secretion
of saliva with a corresponding alimentary motor reaction is produced at
about the thirtieth minute. Sometimes this reaction occurs exactly at the
thirtieth minute, but it may be one or two minutes late. In the interval
there is not the least sign of any alimentary reaction, especially if the
routine has been repeated a good number of times. When we come to
seek an interpretation of these results, it seems pretty evident that the
duration of time has acquired the properties of a conditioned stimulus.

What is the physiological meaning of these time intervals in their role
as conditioned stimuli? . . . Time is measured from a general point of view
by registering different cyclic phenomena in nature, such for instance as
the rising and setting of the sun or the vibration of the pendulum of a clock.
But many cyclic phenomena take place inside the animal’s body.

In a word, the explanation of the striking outcome of Dr.
Feokritova’s experiment seems to be that each feeding of the dog
initiated the stable internal cycle of digestion which activated
somewhat different receptors at each of its phases. The receptor
discharges released by the phase reached after 30 minutes of
digestion were naturally conditioned to the salivary process evoked
and reinforced by each subsequent feeding. The afferent process
conditioned under such circumstances would not be a perseverative
stimulus trace but, rather, the afferent impulse arising directly
from receptor discharges. For this reason considerable confusion
might be avoided if reactions conditioned directly to stimuli aris-
ing from such physiological cycles were called cyclic-phase condi-
tioned reactions.

A cyclic-phase conditioned reaction involving another type of
physiological cycle may easily be set up ( 4 , pp. 417-418). An
electric shock may be delivered to a subject 30 times at regular
intervals of a half minute. Each shock will cause the subject to
react rather strongly, releasing various endocrine secretions and
otherwise upsetting his equilibrium. Presumably the body at once
will begin to shift back to normal in much the same way after
each shock. Presumably also, each phase of this recovery cycle
will activate a somewhat different set of receptors. Thus just
before the onset and cessation of each shock the same set of recep-
tors will be discharging as on the previous occasions and so will
become conditioned to the reactions evoked by the shock, e.g., the
galvanic skin reaction. It naturally follows that if the shock is
omitted the discharge of these receptors will evoke the reaction at
about its usual time of incidence, much as if the shock were deliv-

b h* and stimulus-response ASYNCHRONISM 175

ered. A record of such a cyclic-phase conditioned reaction is repro-
duced as Figure 40. 1

Let it be supposed that a stimulus incidentally becomes con-
ditioned to a strong reaction, such as the one to shock, which in-
volves a certain time for the return of the body to equilibrium.

Fio. 40. Reproduction of a record showing a cyclic-phase conditioned
reaction in a human subject. The tracing shows the reactions to the last
four of 30 induction shocks delivered at 385-second intervals. Presumptive
conditioned galvanic reactions to the temporal interval appear at X, Y, and Z.
The vertical lines have been drawn to show points of simultaneity on the
several tracings. (Reproduced from 4, P- 418.)

Such a stimulus will necessarily evoke the reaction during the
learning process, thereby initiating an internal behavior cycle
which may be some minutes in length. A conditioned stimulus
combination of this nature would yield a conditionable process
which would extend far beyond the range of a true perseverative

1 Under the author's direction, Mr. R. O. Rouse performed a modified
form of this experiment, in which a shock was delivered every 30 seconds.
Seven of eleven subjects displayed temporal conditioning in one form or
another analogous to that shown in Figure 40, some giving evidence of anti-
cipatory or anxiety tendencies. One subject who gave clear indications of
“temporal” conditioning admitted, upon subsequent questioning, that he
had counted; none of the other subjects reported having done this.

1 76

PRINCIPLES OF BEHAVIOR

receptor discharge, and might easily give rise to what would super-
ficially appear to be trace conditioned reactions in which the
stimuli would be separated by many times three seconds. This
presumptive mechanism may explain the paradox that certain
studies, such as those of Warner (11) and Yarbrough (14), purport
to show reactions conditioned to perseverative stimulus traces in
which the supposed “trace” at the point conditioned may be as
old as twenty seconds.

SUMMARY

Numerous experiments have shown that, the gradient of rein-
forcement remaining constant, the most favorable temporal ar-
rangement for the delivery of the conditioned and the uncondi-
tioned stimuli is to have the latter follow the former by something
less than a half second. But as the asynchronism of the onset of
the two stimuli deviates from this optimal relationship in either
direction there is a falling off in the habit strength which will
result from a given quality and number of reinforcements, the
rate of decline in each direction probably being a simple negative
growth or decay function of the nature and extent of stimulus
asynchronism. The situation where the onset of the unconditioned
stimulus, as well as the reaction and its reinforcement, antedates
the optimal relationship by more than a half second or so includes
two special cases which are traditionally known as simultaneous
conditioning and backward conditioning. These are believed to be
portions of the same physiological continuum antedating the point
of optimal stimulus asynchronism. This yields the anterior stim-
ulus-asynchronism gradient.

The case where the unconditioned stimulus falls later than the
optimal relationship has been studied somewhat more than the one
in which it precedes this point. This develops two posterior gradi-
ents of stimulus asynchronism, depending on whether the condi-
tioned stimulus terminates early or continues on to overlap the
unconditioned stimulus. In both events, habit strength declines
markedly as the delay in the onset of the unconditioned stimulus
increases beyond the optimum, the diminution being a simple decay
function of the amount of delay beyond the optimal relationship.
If the conditioned stimulus persists, the resulting habit is said to
be a delayed conditioned reaction. The rate of fall in habit
strength as a function of increased asynchronism is moderate, and

sHs AND STIMULUS-RESPONSE ASYNCHRONISM

*77

the limit of the decline probably has a value considerably above
zero. This level is believed to represent the status of static ele-
ments in stimulus complexes.

In case the conditioned stimulus is instantaneous the resulting
habit is said to be a trace conditioned reaction , the rate of fall in
habit strength is relatively great, and the limit of decline in the
gradient when experimental artifacts are eliminated probably is
zero. It is doubtful if true trace conditioned reflexes can be set
up when the onset of the unconditioned stimulus follows the termi-
nation of the conditioned stimulus by more than about three
seconds.

Both posterior-asynchronism gradients are tentatively regarded
physiologically as increasing functions of the magnitude or inten-
sity of the temporally contiguous afferent discharges. These in
their turn are believed to be increasing functions of the frequency
of impulses given off by the receptors. The gradient of afferent
discharge intensity in the case of delayed conditioned reactions is
supposed to arise from the continuous action of the conditioned
stimulus upon its receptor; that in the case of trace conditioned
reactions is thought to be a mere perseveration or trace of the
afferent action after the conditioned stimulus which originally
initiated it has ceased to act on its receptor. In the case of the
continued action of a stimulus energy on a receptor, the s presum-
ably consists of the afferent discharge arising in the receptor at a
given instant, plus the perseverative traces arising from the stimu-
lation during preceding instants.

The cyclic-phase conditioned reaction superficially resembles
the true trace conditioned reaction. In such situations the “stimu-
lation” involves, in addition to a mere receptor discharge, the
setting in motion of a major physiological cycle such as that of
digestion or the return to equilibrium after an electric shock. If
the stimuli activated by a particular phase of such a cycle are
regularly conditioned to the reaction in question, this reaction will
later be evoked at that phase of the cycle, and consequently at a
certain time following the onset of a stimulus associated with the
initiation of the cycle. Stimuli evidently originating in the diges-
tive cycle in dogs have evoked conditioned reactions as much as
half an hour after the last preceding significant stimulation. The
results of such “temporal” conditioning have somewhat mislead-
ingly been said to yield “long” trace conditioned reactions.

The considerations put forward in Chapters VI, VII, VIII, IX,

PRINCIPLES OF BEHAVIOR

178

X, and the present one enable us to formulate our fourth primary
principle, law, or postulate:

POSTULATE 4

Whenever an effector activity (r — » R ) and a receptor activity (S — * s)
occur in close temporal contiguity (iCV), and this ' S C T is closely associated
with the dimin ution of a need (G) or with a stimulus which has been
closely and consistently associated with the diminution of a need (G),
there will result an increment to a tendency (A sHr) for that afferent
impulse on later occasions to evoke that reaction. The increments from
successive reinforcements summate in a manner which yields a com-
bined habit strength (sHr) which is a simple positive growth function
of the number of reinforcements (N). The upper limit (m) of this
curve of learning is the product of (1) a positive growth function of the
magnitude of need reduction which is involved in primary, or which is
associated with secondary, reinforcement; (2) a negative function of the
delay (f) in reinforcement; and (3) (a) a negative growth function of
the degree of asynchronism (*') of S and R when both are of brief dura-
tion, or (b), in case the action of S is prolonged so as to overlap the
beginning of /?, a negative growth function of the duration (*") of the
continuous action of S on the receptor when R begins.

NOTES

Mathematical Statement of Postulate 4

The mathematical statement of Postulate 4 is distinctly more concise, con-
venient, and informative than is the verbal formulation given above. It has
several cases depending on whether (1) S is prolonged and overlaps the beginning
of R temporally, and (2) their degree of asynchronism in case they are of brief
duration and do not overlap temporally. As an illustration of the case where R
follows and overlaps the continuous action of S on the receptor for duration t" t
we have the following equation:

b H r = M (1 - e~ k ' B )er it e~ ui ' (1 - e""0, (16)

where,

Af = 100 habs, the physiological maximum of habit strength ;

e = a mathematical constant usually taken in the present work as 10 ;

10 = a constant change in a measurable objective criterion which results in a
need reduction ;

t = the delay in reinforcement;

l> _ Tr — Tk — -66, where S is of more than instantaneous duration and over-
laps the beginning of R ;

Tr ■= the time of the beginning of R;

B Ha AND STIMULUS-RESPONSE ASYNCHRONISM

l 79

T‘ s = the time of the beginning of S;

N = the number of reinforcements ;
k, j, u, and i = empirical constants.

The meaning of equation 16 may be clarified by the following example.
Let it be supposed that we have a simple learning situation in which 20 rein-
forcements are given ( N = 20); that 5 grams of a standard food are given a
canine subject at each reinforcement (to = 5) ; that the reinforcement is given
3 seconds after the ,C r ( t =3); and that S begins a continuous action on the
receptor 2 seconds before the beginning of R ( t 0 = 2). Taking the values of
k, j, u , and t from previously given equations (11, 14, 20, and 6) fitted to empirical
data, and substituting, we have,

3 Hr = 100(1 - 10“ 153 * 5)10- 00672 X 3 X 10 ” -2515(2— .66) (1 _ 1 0 -.018 X 20). (17)

Solving by easy stages, we have,

° H * - H 1 -m) x Tok x 2ik (1 - 10 - 0I8 * 20 >

= 100(.3562)(1 - 10-018x20)

= 35.62(1 - 10- ois x 20) f

which presents the familiar equation expressing the positive growth curve of
simple learning. Solving further, we have,

sHr = 35.62 X .5635
= 20.1 habs.

In case the learning situation is the same as above except that both S and R
are of brief duration, S preceding R by .9 second, the equation becomes :

bHr = 10(1 - 10 - 1 “ x 5)10- 00672 X e X 10-1.182(.9-.44)(1 _ 1Q--013 X 20). (Ig)

flHjj = 12.5 habs.

In case the learning situation is the same as above except that R precedes S
by one-tenth of a second, we have :

bHr = 10(1 - 10 - I 53 * 5)10- 00672 X 3 X 10+1 068<-.l-.44)(l _ 1Q--018 X 20). ( 1 9 )

.*. bHr = 11.0 habs.

The Equation of the Curve of Figure 38

The negative growth function fitted to the values represented by the hollow
circles of Figure 38 is :

y = 62.6 X 10" 2515C*-!) + 30.4 (20)

where y is the per cent of antedating reactions and x is the time from the beginning
of the conditioned stimulus to that of the unconditioned stimulus, the reaction,
and the reinforcement. The number 30.4 represents the asymptote.

i8o

PRINCIPLES OF BEHAVIOR

The Equations of the Curves of Figure 39

The negative growth function fitted to the hollow circles at the right of Figure
39 is:

y = 30.5 X 10 -1 182 C» - - 5 ) + 6.5 (21)

where y is the per cent or probability of conditioned-reaction evocation on the
test trials and x is the time at which the reaction occurs (Tr) less the time at
which the stimulus occurs (Ts ) ; in case R precedes S, x will, of course, be nega-
tive. The number 6.5 represents the asymptote.

The negative growth function fitted to the values represented by the solid
circles at the left side of Figure 39 is :

y = 22.5 X 10 -i <*«< 25^) + 65> (22 )

in which y and the value 6.5 mean the same as in the equation fitted to the right-
hand gradient, and the values of x must be less than .44 second.

Probably a somewhat more general and significant manner of writing the
above equations is as follows :

V = 35.9 X 10 - *' - 6.5, (23)

in which,

P = the per cent or probability of conditioned-reaction evocation;
t* = T R -Th - .44;

Tr = the time of the instantaneous occurrence of R in seconds;

T's = the time of the instantaneous occurrence of S in seconds;

v = — 1.068 if C is negative, but + 1.182 if t" is not negative.

Ultimately of course, the values of p in equations 20, 21, 22, and 23 will need to
be converted into units of amount on the basis of the normal probability function
(see p. 311 ff.), which will presumably change the equations in question as well as
the forms of Figures 38 and 39. Despite this defect it is believed that these
equations and figures have a certain amount of expository value as well as sug-
gestive significance for further developments.

The Stimulus- Asynchronism Gradients and the Parameters of the Curve

of Habit Acquisition

The question of the relation of the three stimulus-asynchronism gradients of
conditioning to the slope and asymptote of the curve of habit-strength acquisition
as a function of the number of reinforcements, arises here just as we have seen
parallel questions arise in connection with the amount of the reinforcing agent
employed and the gradient of reinforcement. It is clear that a satisfactory
theory of learning and of learned behavior requires a knowledge of this relation-
ship in all the cases mentioned. Unfortunately no relevant evidence has been
found which is sufficient to yield a decisive indication as to the relationship in
the case of the stimulus-asynchronism gradients. However, largely in order to
pose the question before students of behavior principles, we postulate that it is
one of several determinants of the asymptote of habit strength (m) with un-
limited reinforcements.

bHs AND STIMULUS-RESPONSE ASYNCHRONISM

181

The Derivation of the Equation Expressing Postulate 4

It is assumed that,

M' = M (1 - «-*"),

(12)

m’ = M'e"*,

(14)

either

m = m'e~ ut

(24)

where (20)

or (23)

t' = T r - Ts ~ .66

m — m'e~ vt '

(25)

and

s Hr = m( 1 - 10-"*)

(26)

where M, m', and m are respectively : the absolute upper physiological limit
of habit strength with unlimited reinforcement, the upper limit of habit strength
as determined by the nature and amount of the reinforcing agent employed, the
limit of habit strength as determined by the delay in reinforcement, and the limit
of habit strength as determined by the degree of receptor-effector asynchronism.

Now, substituting equation 24 (or 25) in equation 26, we have,

8 H r = - 10“"*) (27)

Substituting equation 14 in equation 27, we have,

a H R = M’er»e-*(X - lOr*") (28)

Substituting equation 12 in equation 28, and recalling that M = 100 habs, we
have,

s Hr = 100(1 - e~ kv )e~ it e-^(l - 10""*),

which is one of the alternative equations sought (16).

It is altogether probable that there are other important factors which enter
into the determination of habit strength in simple learning or conditioning situa-
tions. Two of these are the intensity of the conditioned stimulus ( S) and the
vigor or intensity of the reaction ( R ). Specific researches analogous to the studies
of Gantt, Perm, Williams, Hovland, and others cited in the preceding pages need
to be performed on these probable factors before their relationship to habit
strength can be postulated with much confidence. In the end a grand investiga-
tion involving the finding of the joint reaction potentiality of all the presumptive
determining factors when taken in all combinations of the representative values
of each (after the manner of Perm's study of habit strength and motivation, 8)
must be carried out before a really dependable equation for Postulate 4 can be
written. The point is that habit strength as a function of t when id and t' are
constant at a certain value, is not necessarily the same as it will be when w and t'
are constant at other values. This will be a huge task, but the outcome should
be worth the labor involved. It seems unlikely that the Fisher-design type of
experiment will yield dependable indications of the complex hyperspatial curva-
tures which almost certainly will be found. Meanwhile, the equations given
above may serve as points of departure for further empirical analyses.

1 82

PRINCIPLES OF BEHAVIOR

REFERENCES

1. Adrian, E. D. The basis of sensation. New York: W. W. Norton and

Co., 1928.

2. Graham, C. H. Vision: III. Some neural correlations. Chapter 15 in

Handbook of general experimental psychology, C. Murchison, editor.
Worcester, Mass.: Clark Univ. Press, 1934.

3. Guthrie, E. R. Conditioning as a principle of learning. Psychol. Rev .,

1930, 87, 412-428.

4. Hull, C. L. Learning: II. The factor of the conditioned reflex. Chap-

ter 9 in Handbook of general experimental psychology, C. Murchison,
editor. Worcester, Mass.: Clark Univ. Press, 1934.

5. Kappauf, W. E., and Schlosberq, H. Conditioned responses in the white

rat. III. Conditioning as a function of the length of the period of
delay. J. Genet. Psychol., 1937, 60, 27-45.

6. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

7. Pavlov, I. P. Lectures on conditioned reflexes (trans. by W. H. Gantt).

New York: International Publishers, 1928.

8. Perin, C. T. Behavior potentiality as a joint function of the amount

of training and the degree of hunger at the time of extinction. J.
Exper. Psychol., 1942, 30, 93-113.

9. Rosenblueth, A. Central excitation and inhibition in reflex changes of

heart rate. Amer. J. Physiol., 1934, 107, 293-304.

10. Switzer, S. A. Backward conditioning of the lid reflex. J. Exper.

Psychol., 1930, IS, 76-97.

11. Warner, L. H. The association span of the white rat. J. Genet. Psychol.,

1932, 41, 57-90.

12. Wolflb, H. M. Time factors in conditioning finger-withdrawal. J. Gen.

Psychol., 1930, 4, 372-379.

13. Wolfle, H. M. Conditioning as a function of the interval between the

conditioned and the original stimulus. J. Gen. Psychol., 1932, 7, 80-103.

14. Yarbrough, J. U. The influence of the time interval upon the rate of

lea rn i n g in the white rat. Psychol. Monog., 1921, SO, No. 135.

CHAPTER XII

Stimulus Generalization

The preceding chapters have shown that learning takes place
according to various principles of reinforcement. In giving this
account we have followed the conventional practice of character-
izing learning as the setting up of receptor-effector connections.
Moreover, we have represented these connections by such symbols

as sH r , 8 H r , and S >s >r >R, which specify only the

receptor and effector processes actually involved in the reinforce-
ment. It is now necessary to point out that while there is every
reason to believe that each reinforcement does result in the con-
nection represented by the symbolism, the actual outcome is much
more complex than this. The fact is that every reinforcement
mediates connections between a very great number of receptor and
effector processes in addition to those involved in the reinforcement
process and represented in the conventional symbolism gH R . Sev-
eral groups of such additional and indirectly established receptor-
effector connections may be distinguished:

1. The reaction involved in the original conditioning becomes con-
nected with a considerable zone of stimuli other than, but adjacent to,
the stimulus conventionally involved in the original conditioning; this is
called stimulus generalization.

2. The stimulus involved in the original conditioning becomes con-
nected with a considerable zone of reactions other than, but related to,
the reaction conventionally involved in the original reinforcement; this
may be called response generalization. 1

3. Stimuli not involved in the original reinforcement but lying in a
zone related to it become connected with reactions not involved in the
original reinforcement but lying in a zone related to it; this may be called
stimulus-response generalization.

The present chapter is particularly concerned with certain phe-
nomena of stimulus generalization and some of their implications
concerning adaptive behavior.

1 The present analysis indicates that response generalization is a rather
complex secondary phenomenon; space is not available in the present work
for an adequate treatment of it.

183

184

PRINCIPLES OF BEHAVIOR

PRIMARY STIMULUS-QUALITY GENERALIZATION

The molar principle of primary stimulus generalization is now
well established both qualitatively and quantitatively, mainly by
conditioned-reaction experiments. A few of the indirectly acquired

Fia. 41. Diagram of re-
ceptor-effector convergence
arising from the primary
stimulus generalization set
up concurrently with the
conditioning of b Hr. S h S 2t
S a represent positions at
progressively greater dis-
tances on one side of S 0 on
a one-dimensional stimulus
continuum, and Si', S*, S*
represent corresponding po-
sitions on the other side of
So on the same stimulus
continuum. In this notation,
So represents the stimulus
when considered as involved
in the reinforcement proc-
ess, and So, Si, S 2 , S,, etc.,
stimuli when considered as
evoking reaction. S 0 and S Q
are understood to fall at the
same point on the stimulus
continuum.

receptor-effector connections set up by
the conditioning of S 0 to R are repre-
sented in Figure 41 as originating in the
S’s with integral subscripts. Since all
of these potentialities of reaction evoca-
tion converge from different stimulus
possibilities upon the same reaction,
stimulus generalization may be said to
generate a receptor-effector convergence.

Under certain circumstances, e.g., in
long trace conditioned reactions, gener-
alization normally extends into receptor
modes other than that involved in the
reinforcement. Thus Pavlov reports the
case of a defensive salivary reaction in
a dog conditioned to the trace of a tac-
tile stimulus, the reaction in question
subsequently being evoked by a thermal
stimulus of 0° centigrade (11, p. 113).
It is important to note that while pri-
mary stimulus generalization may pass
the boundaries of a given sense mode,
this is not usual. Stimuli conditioned to
one sense mode will ordinarily generalize
only to other stimuli in the same sense
mode, e.g., from one auditory vibration
rate or intensity to another, or from one
light wave length or intensity to another.
In general the more remote on the stim-

ulus continuum the evoking stimulus (S) is from that originally

conditioned ( S ) , the weaker will be the reaction tendency mobilized

by it.

The quantitative law of primary stimulus generalization is
nicely illustrated by an experiment reported by Hovland ( 4 )• This
investigator conditioned the galvanic skin reaction in human sub-
jects to a pure tone, e.g., of 1,967 cycles per second, and then meas-

STIMULUS GENERALIZATION

185

ured the amplitude of the reaction evoked at this pitch and at three
other pitches separated from each other by an equal number of
discrimination thresholds (25 j.n.d.’s). The pooled results from
twenty subjects are shown in Figure 42. There it may be seen
that:

1. The amplitude of galvanic skin reaction diminishes steadily with the
increase in the extent of deviation (tf| of the evocation stimulus (S) from
the stimulus originally conditioned (S).

200 150 100 50 0 50 100 150 200

NO. Of JUST NOTICEABLE DIFFERENCES DISTANT FROM POINT OF REINFORCEMENT (D)

Fia. 42. Empirical generalization gradient of conditioned galvanic skin
reaction derived from data published by Hovland (4). Note that the gradient
extends in both directions on the stimulus continuum (vibration rate) from
the point originally conditioned.

2. This diminishing generalization tendency extends symmetrically in
both directions along the stimulus dimension.

3. The quantitative course of the diminution in the generalization
tendency approximates rather closely a negative growth function of the
amount {d) that S deviates from & as measured in discrimination thresh-
olds (jm.d.’s). This is attested by the closeness with which the smooth
curves, representing a simple decay function fitted to the generalization
data, approximate the circles of Figure 42.

1 86

PRINCIPLES OF BEHAVIOR

4. The asymptote or lower limit of the generalization gradient falls
at 12.3 millimeters, rather than at zero. The failure of the gradient to
approach zero as a limit is regarded as an experimental artifact due in
part to the fact that previous to conditioning this reaction is evokable
in appreciable amounts by any stimulus of even moderate intensity, in
part to sensitization, and in part to the reaction becoming conditioned
somewhat to the static stimuli arising from the experimental environment.
The environmental portion of the stimulus situation, of course, remains
constant throughout the changes in the tonal stimulus, which alone pro-
duce the gradient. Accordingly it is concluded that the asymptote of the
* true generalization gradient is probably zero (5).

PRIMARY STIMULUS-INTENSITY GENERALIZATION

r a second study ( 5 ) employing the same general apparatus
arrangement, Hovland attempted to determine the quantitative

i ^'“Pkical stimulus-intensity generalization gradient of the con-
ditioned galvanic skin reaction plotted from data published by Hovland (5).
Note that while the gradients slope downward with increasing degrees of
deviation from the point originally conditioned, the steepness of the slope i s
distinctly less than that of the stimulus-quality generalization shown in
Figure 42.

law of stimulus-intensity generalization. The procedure was to
condition a given intensity of a simple sinusoidal sound wave to
the galvanic skin reaction, and then test other intensities of the
same frequency at 50 j.n.d. intervals. While the results of general-

STIMULUS GENERALIZATION 187

ization were seriously complicated by specific effects of intensity
which were quite independent of generalization, it is believed that
these latter effects were substantially eliminated by pooling the
results of two groups of subjects, one in which the reaction ampli-
tude was determined for intensities greater than that conditioned
and one in which the determination was made for intensities which
were less. The outcome of this experiment is shown by the circles
in Figure 43. A negative growth function has been fitted to these
values; this is represented by the smooth curve running among the
circles. The fit is reasonably good.

It is evident from an inspection of Figure 43 that stimulus
intensity also manifests a generalization gradient, but that the rate
of fall of the gradient per j.n.d. of deviation from the point con-
ditioned is distinctly less than is that for stimulus-quality general-
ization as shown in Figure 42. The latter has a fractional rate
of decremental change per j.n.d. of deviation from the point con-
ditioned of approximately 1/33, whereas the former has an F-value
of approximately 1/77.

THE CONCEPT OF EFFECTIVE HABIT STRENGTH (bHr)

From the foregoing it is evident that the simple notion of
habit strength, as indicating merely the strength of connection
between the stimulus and the reaction involved in the original
reinforcement process, must be radically expanded before the in-
fluence of learning on functional activity is to be understood and
represented in a realistic manner. It is true that the various prin-
ciples of reinforcement when perfected will presumably enable us
to predict with precision the strength of the connection between
the conditioned stimulus and the associated reaction. This is all
right so far as it goes, but it represents only a small portion of
the zone of reaction evocation potentialities set up by a given
reinforcement. The strength of the connections at the other points
of the zone can be determined only from a knowledge of the strength
of the receptor-effector connection (bHr) P°^ n ^ reinforce-

ment and the extent of the difference (d) between the position of
the conditioned stimulus (S) and that of the evocation stimulus (5)
on the stimulus continuum connecting them. Thus there emerges
the concept of functional or effective habit strength, which we shall
represent by the symbol sHn. This symbol will be used to desig-
nate the habit strength throughout the entire zone of habit forma-

PRINCIPLES OF BEHAVIOR

188

tion which is set up by a given reinforcement process, or, with modi-
fications ( ssHr , according to conditions), the summation of the
effects of two or more reinforcement processes. The symbol b Hr
will be reserved, as before, to indicate the strength of habit at the
point of reinforcement; i.e., when d = zero,

sH r — sHr*

THE CONCEPTS OF STIMULUS DIMENSION AND AFFERENT

GENERALIZATION CONTINUUM

It is clear from the now familiar causal relationship S >

b— >R that in the evocation process (1) the stimulus

energy (5) determines which receptor shall be activated and the
occasion of its activation, (2) the nature of the receptor thus
activated determines the detailed characteristics of the receptor
discharge (s), and (3) the reaction (f?) is only indirectly a func-
tion of S by virtue of the fact, and only to the extent, that s stands
in a one-to-one relationship to S. There is reason to believe that
this parallelism is never exact and that certain factors such as
afferent neural interaction may produce marked deviations.

These considerations have definite implications for certain phe-
nomena of stimulus generalization. Thus it is clear that there can
be no primary stimulus generalization unless there is some parallel
physical variability in the stimulus energy to serve as its basis;
for example, there could be no generalization in the stimulus dimen-
sion of frequency or amplitude of sound waves if sound waves did
not present such dimensions of variability. Secondly, generaliza-
tion cannot take place on a given stimulus dimension if the relevant
receptor does not respond differentially to variability in that dimen-
sion; for example, organisms which are color blind, i.e., those whose
light receptors do not yield differential responses to variations in
wave length of light, can hardly be expected to show a generaliza-
tion gradient along this stimulus dimension. Accordingly there
emerge the contrasted concepts of stimulus dimension and afferent
generalization continuum , the latter being the differential afferent
response (s) corresponding in varying degrees to variation in a
given stimulus continuum. With few exceptions the receptors of
normal higher organisms appear to yield afferent generalization
continua for all the physical stimulus dimensions to which they
respond at all. Conversely, for every empirically observed 'primary

STIMULUS GENERALIZATION 189

generalization dimension, some measurable dimension of the physi-
cal situation has usually been found.

Since most stimulus energies vary in more than one dimension,
it comes about that any given bit of learning is likely to set up
generalization gradients along several continua simultaneously. In-
asmuch as the relevant stimulus dimensions may vary independ-
ently, the resulting mixture of several generalization gradients in a
single generalization situation often greatly complicates the inter-
pretation of experimental results involving generalization. This is
particularly true in the field of vision, where there may be com-
bined the stimulus dimension corresponding to white light and the
innumerable afferent generalization continua arising from the
simultaneous combinations of two or more wave lengths. To com-
plicate matters still further there is the neural interaction (p. 42)
of processes going on in different parts of the retina, which makes
the afferent discharge ( 3 ) of a given retinal element not merely a
function of the stimulus energy (S) impinging on it but also of the
energies S', etc., impinging on neighboring elements at or about
the same time. This complication becomes especially apparent in
figured or spatially patterned stimulus situations in which there
may emerge such generalization continua as degrees of curvature
or angle of outline, size of figure, brightness, contrast or difference
between portions of the area stimulated, the angle of rotation of
the figure, and so on. When numerous physical dimensions are
mixed in various ways and, particularly, where interaction occurs
between different parts of the retina, the nature and amount of the
generalization effects are extremely difficult to predict, as the exten-
sive experimental investigations of Lashley have shown (10). There
is reason to hope, however, that these problems will finally yield
to the joint and systematic study of primary generalization gradi-
ents and gradients of afferent neural interaction. Certain investiga-
tions of the Gestalt psychologists should prove valuable in clarify-
ing the latter type of problem.

IN WHAT UNITS SHALL THE DIFFERENCE BETWEEN THE CONDI-
TIONED STIMULUS AND THE EVOCATION STIMULUS IN
PRIMARY GENERALIZATION BE MEASURED?

Discrimination investigations show with considerable clarity that
while s is a function of S, the relationship is usually by no means
linear, and frequently it is not even monotonic. A well-known

190

PRINCIPLES OF BEHAVIOR

example of the latter type of irregularity concerns the octave in
auditory wave frequency. There is a tendency for generalized re-
actions to be evoked more strongly by stimuli which are even
multiples of the frequency originally conditioned than by certain
intermediate rates (9 ) , a fact which is in conflict with the prin-
ciple represented by Figure 42. The present a jpriori unpredictability
in the character of many receptor responses as a function of the
several stimulus dimensions is presumably due to our ignorance
regarding the physiology of the receptors. Because of these irregu-
larities it will probably be impossible to represent all generalization

gradients as any uniform function of the stimulus dimensions in-
volved.

There remains, however, the possibility of expressing general-
ization gradients in terms of distances on the afferent generalization
eontinuum. The natural unit of measurement on this continuum
is the discrimination threshold, or j.n.d. This is a difference be-
tween two stimuli on a given stimulus dimension (the other dimen-
sions remaining constant) such that at the limit of discrimination
training the organism will consistently give differential reactions
to the two stimuli on 75 per cent of the trials. Presumably because
the process of discrimination involves the joint effect of variability
in the stimulus dimension and the corresponding afferent reaction
of the receptor, the generalization gradients in general appear to
be rather accurately and simply expressible as negative growth

functions of the stimulus dimension when transmuted into j.n.d.
units.

GENERALIZATION BY MEANS OF IDENTICAL STIMULUS

COMPONENTS

There may now be mentioned a second form of stimulus gen-
eralization, that which arises indirectly because conditioned stimuli
are not simple but are normally compounded of the simultaneous
discharge of a very great number of distinct receptors. Let it be
supposed, for example, that a salivary reaction has been condi-
tioned to a compound stimulus consisting of a group of sound waves
produced by an organ pipe (5 0 ) and a group of light waves pro-
duced by an incandescent filament ( S v ), and that each of the two
stimulus aggregates has independently acquired a superthreshold
potentiality for evoking the reaction. Now, if a second stimulus
situation consisting of the vibrations produced by the organ pipe

STIMULUS GENERALIZATION 19 1

(S a ) were presented either alone or in combination with a cutaneous
vibration (<S C ), say, the reaction would be evoked, owing to the
presence in the second stimulus situation of the originally condi-
tioned auditory component (S a ). The operation of this principle
has been investigated by the author at a gross molar level in con-
nection with an experimental study of generalizing abstraction or
concept formation (£) ; it has also been utilized extensively by
Thorndike (13) in the explanation of certain forms of training
transfer.

It is tempting to assume with Guthrie (£) that all primary
generalization is built on this model. Involved in such an hypoth-
esis there is the implicit assumption that the afferent discharge
initiated by every stimulus energy consists of a large number of
afferent “molecules,” and that the contiguous receptor discharges
on any given stimulus continuum have a considerable portion of
their afferent molecules in common while differing with respect to
certain others. Thus one receptor discharge might consist of the
molecules a, 6, c, d, e, /, g, whereas the adjacent one on the afferent
generalization continuum would consist of the molecules b, c, d, e,
/, g, h, the next one would consist of c, d, e, /, g, h, i, and so on.
One discrimination threshold on such a continuum would be the
physical measure of the qualitative or quantitative variation in the
stimulus energy which would change enough afferent molecules so
that a subject, at the limit of training, would react differentially
to the two stimuli on 75 per cent of the trials. It is quite possible
that something of this nature will turn out to be the ultimate
physiological explanation of primary generalization. As yet, how-
ever, proof is lacking on the molecular level, and there seems no
immediate prospect of securing a critical test of the hypothesis.
Meanwhile we must get along as best we can with a molar analysis
based on empirically determined functional relationships, e.g., those
presented graphically in Figures 42 and 43.

SECONDARY STIMULUS GENERALIZATION

Except under certain special circumstances, such as those of
sensitization (7, p. 431) or “long” trace conditioned reactions, con-
ditioned stimuli probably do not show generalization into other
receptor modes. Yet we may recall the name of a person with
about equal probability on seeing either his face or the back of
his head, at the sound of his voice or even his footstep. Such be-

I 9 2

PRINCIPLES OF BEHAVIOR

havior presumably comes about not through primary stimulus gen-
eralization, but indirectly through each reaction being specifically
learned. For example, a salivary reaction may be conditioned to
a tactile vibration, to an auditory vibration, and to a flash of light.
Since each of the three stimuli leads indifferently to the salivary
reaction, this type of habit organization constitutes a learned recep-
tor-effector convergence, quite distinct from the convergence pro-
duced by primary gen-
eralization (Figure 41).
Such a learned converg-
ence is represented dia-
grammatically in Figure
44.

Receptor-effector con-
vergence is of particular
importance in behavior
theory, since it appears
to be a medium of the
automatic transfer of
training effects. It is significant in the present context because,
as a special case of such habit transfer, it seems to mediate what is
known as secondary stimulus generalization.

An apparent case of secondary stimulus generalization has been
reported by Shipley (12) and verified by Lumsdaine (3, p. 230).
These investigators presented a subject with a flash of light fol-
lowed by the tap of a padded hammer against the cheek below the
eye, thus conditioning lid closure to the light flash. Next, the same
subject was repeatedly given an electric shock on the finger. This
evoked not only a sharp finger withdrawal from the electrode, but
lid closure as well. Finally the flash of light was delivered alone.
It was found in a considerable proportion of the subjects of both
experiments that during this latter manoeuvre the light evoked
finger retraction even though the former had never been associated
with either the shock or the finger retraction. The interpretation
is that the light evoked the lid closure, and the proprioceptive
stimulation produced by this act (or some other less conspicuous
act conditioned at the same time) evoked the finger retraction.

Lumsdaine’s photographic records (3, p. 231) of the process
tend to support the view that in this experiment the wink reaction
served as a mediating agent, since they show that, typically, when
the light evoked finger retraction the lid closure usually took place

TOUCH

SOUND

SALIVATION

LIGHT

Fig. 44. Diagram of a specifically learned
convergent excitatory mechanism. Each stimu-
lus is assumed to have been conditioned to the
salivary reaction on a separate occasion.

STIMULUS GENERALIZATION

between the flash of the
light and the finger
movement. Occasionally,
however, the two reac-
tions occurred at the
same time, and some-
times the finger move-
ment even preceded the
blink. This, of course,
could not have happened
if the finger retraction
was evoked by the pro-
prioceptive stimuli aris-
ing from the lid closure.
It is possible, however,
that numerous other re-
actions were conditioned
at the same time as the
wink, and that proprio-
ceptive stimuli from all
of them became condi-
tioned to finger retrac-
tion. If occasionally the
lid closure should have
occurred later than the
other reactions, the pro-
prioception from the lat-
ter might easily have
evoked the finger retrac-
tion alone. While these
considerations compli-
cate the interpretation
of Shipley’s results to
the extent that they do
not constitute an un-
equivocal proof of the
mechanism of secondary
generalization, the ex-
periments do demon-
strate the existence of
secondary generalization

PARI A

LIGHT ^

SHOCK

WINK

PART B

SHOCK

^WINK

RETRACT— i-P

Ret

PART C

LIGHT —WINK P w RETRACTION

Fia. 45. Diagrammatic representation of the
evolution of secondary stimulus generalization
in the Shipley-Lumsdaine experiments. Part
A shows the basic convergent mechanism, the
light-wink portion having been set up by
means of a previous conditioning process.
Part B represents the conditioning of the pro-
prioceptive stimuli of two reactions evoked by
a shock, each to the other reaction through
simultaneous occurrence closely associated with
reinforcement, i.e., cessation of the shock. Part
C shows the final indirect generalization. The
light evokes the wink (from Part A) ; the wink
produces a proprioceptive stimulation ( Pw
from Part B ) ; and Pw evokes a finger retrac-
tion as conditioned in Part B. Thus the finger
retraction has been indirectly generalized from
the shock to the light through the mediation
of the wink reaction upon which both light
and shock as stimuli converge. Throughout
this diagram the arrows with solid shafts repre-
sent receptor-effector connections which were
in existence at the outset of the learning proc-
ess here under consideration, and the arrows
with broken shafts represent connections set
up during the experiment.

194 PRINCIPLES OF BEHAVIOR

and at the same time offer a convenient illustration of a plausible
explanatory mechanism. This is shown diagrammatically in Figure
45, the legend of which gives a somewhat detailed explanation.

It is evident that when to the complexities of primary general-
ization mentioned earlier in the chapter there are added those natu-
rally arising from secondary generalization (p. 191), the task of
predicting generalization effects becomes almost hopeless because
secondary generalization is so largely dependent on fortuitous ele-
ments in the history of the individual organism, and these are
usually not known to the investigator. In this connection we may
note the great facility of normal human beings in the acquisition
and use of speech reactions and the recent experimental evidence
that speech reactions operate in subtle ways to mediate secondary
generalization (1). Because of these considerations, the results of
introspective or verbal reports of the existence of generalization
continua which do not conform with a reasonable approximation to
some objective stimulus continuum in situations where interaction
effects are presumably not marked are open to a certain amount of
doubt. When uncertainty arises as to the status of such dimen-
sions, the situation should be clarified by a comparison of the
results of introspection with the generalization gradients produced
in naive organisms presumably lacking the mediating speech habits.

An important conclusion flowing from the preceding considera-
tions is that the common-sense notion of similarity and difference
is based upon the presence or absence of primary generalization
gradients, whereas so-called logical or abstract similarities and dif-
ferences 1 arise from secondary, learned, or mediated similarities
and differences, particularly those mediated by verbal reactions.

THE “STIMULUS-LEARNING” AND “STIMULUS-EVOCATION”

PARADOXES AND THEIR RESOLUTION

The conventional representation of learning as the formation of
simple bonds gives rise to certain paradoxes. The flux of the world
to which organisms must adapt has infinite variety, and therefore
stimuli, especially conditioned stimuli, are never exactly repeated.
But superthreshold (adaptive) reaction potentials (p. 326 ff.)

1 Consider, for example, the similarity among weapons: this lies hardly
at all in the receptors activated. Other examples in point are: similarity
or difference in degree of value, weight, height, etc., as represented by num-
bers; as primary stimuli, 10 and 90 are hardly as different as are 10 and
12, yet 90 — 10 is forty times as great as is 12 — 10.

STIMULUS GENERALIZATION

*95

usually require more than one reinforcement to be raised above the
reaction threshold. Since the stimuli are not exactly repeated, how
can more than one reinforcement occur? This is the stimulus-
learning paradox. But even if a superthreshold bond should be
established, it becomes a mystery how it could ever evoke a reac-
tion at a time of need, because the exact stimulus would probably

Fio. 46. Graph showing how the subthreshold primary stimulus generaliza-
tion gradients from five distinct points on a stimulus continuum theoretically
may summate to superthreshold values not only at the points of reinforce-
ment but at neighboring points which have not been reinforced at all. Solid
circles represent the results of single reinforcements; hollow circles represent
the results of summation. The reaction threshold is arbitrarily taken at 5.
Note that the major reaction tendency accumulates especially at the mid-
point of the distribution of stimulated points on the continuum, but that
superthreshold reaction potentialities extend beyond the range of the points
conditioned.

never again be encountered. This is the stimulus-evocation para-
dox. The principle of primary stimulus generalization now avail-
able enables us to resolve both of these paradoxes.

Let it be supposed, in a particular reinforcement situation in
which an effective habit strength of five habs is a minimum neces-
sary to evoke reaction, that each reinforcement connects the con-
ditioned stimulus to the reaction with a strength of three habs;
that five reinforcements occur, the five conditioned stimuli involved
falling on the same stimulus continuum at uniform intervals of

PRINCIPLES OF BEHAVIOR

196

10 j.n.d.’s; that for a given potential stimulus each j.n.d. of addi-
tional deviation on the stimulus continuum from the point condi-
tioned decreases the effective habit strength by approximately one
thirty- third ; and that the several habit strengths thus active at a
given point on the stimulus continuum summate to produce a joint
habit strength, as would the number of reinforcements necessary
to produce each if they were to be given in some standard rein-
forcement sequence (Postulate 4).

The dynamics of this supposititious situation are represented
diagrammatically in Figure 46. The habit strengths of primary rein-
forcement are shown by the five solid circles, the generalization
gradients of each being indicated by the negative growth curves
sloping downward in two directions. From an inspection of these
overlapping gradients it is evident that in addition to the three
habs arising from the reinforcement at a given point, there are to
be combined four lesser generalization values. A little further
study will show that the nearer to the middle of the distribution
of conditioned stimuli a point of reinforcement stands, the larger,
upon the whole, will be the generalization values to be combined.

Combining these five 3 H R values at each of the five points of rein-
forcement and at intervals of 5 j.n.d.’s on either side, there is
obtained the series of summation values represented by the upper
curve drawn through the five hollow circles. An inspection of the
latter curve discloses the following:

I. A number of subliminal reinforcements conditioning the same
reaction to distinct stimuli closely spaced along a stimulus con-
tinuum may yield an unbroken superthreshold zone of habit
strengths extending well beyond the range of the conditioned stimuli
in question.

II. The point of maximum habit strength tends to fall at the
middle of the distribution of conditioned stimuli.

III. Points on the stimulus continuum between two points of
reinforcement, but themselves not reinforced at all, have an effec-
tive habit strength only a little less than the mean of the strengths
of the adjacent reinforcement points.

IV. Points on the stimulus continuum falling beyond the range
of the stimuli involved in the conditioning process also rise above
the reaction threshold but in progressively smaller amounts the
more remote they are from the central tendency of the distribution

of the stimuli conditioned.

STIMULUS GENERALIZATION

*97

By the first item in the above summary, isolated stimuli sub-
liminally reinforced only a single time ultimately become supra-
liminal through the summation of effective habit strength (ssH R )
generated at the point in question with that generalized from other
points of reinforcement; thus is resolved the stimulus-learning
paradox. By the summation of generalized effective habit strengths
from adjacent stimulus points of reinforcement, supraliminal habit
strengths evolve at points which have never been reinforced at all;
thus is resolved the stimulus-evocation paradox.

SUMMARY

Under favorable experimental conditions in a learning situation
both the conditioned and the unconditioned stimuli may be held
relatively constant. In this way a connection is said to be set
up in the nervous system between the afferent discharge (s) aroused
by the conditioned stimulus (S) and the efferent discharge (r)
which leads to the reaction ( R ). Actually, however, very much
more than this results; the reaction is conditioned not only to a
tone (5) but to a whole zone of tones of other pitches and inten-
sities spreading in both directions along each dimension from the
point conditioned. All of these stimuli are functionally equivalent
in that they have the capacity to evoke the same reaction. This
spreading of the results of learning to other stimuli is called pri-
mary stimulus generalization. The fact that many stimuli alike
possess the potentiality of evoking the same reaction constitutes
primary stimulus equivalence.

Experiments show, however, that the strength of the habit gen-
eralized to stimuli other than the one originally conditioned dimin-
ishes progressively as the difference between S and S increases.
When the magnitude of this difference is measured in units of the
discrimination threshold (j.n.d.), the gradient of generalization
closely approximates a simple negative growth or decay function.

The introduction of the phenomenon of primary stimulus gen-
eralization makes it quite clear that knowing the habit strength
at the approximate points of reinforcement is now sufficient to
enable us to predict the reaction potentiality, motivation (drive)
remaining constant. The actual or effective habit strength mobiliz-
able by a given evoking stimulus (S) is a joint function of the
habit strength at the point or points reinforced and the difference

PRINCIPLES OF BEHAVIOR

198

on their generalization continuum between the point or points of
reinforcement and the stimulus point of evocation. There thus
emerges the necessity for a new symbolic construct, that of effective

habit strength («//«).

The concept of “stimulus dimension” may be contrasted with
that of “afferent generalization continuum.” The first of these
expressions refers to the physical characteristics of the stimulus
energy; the second, to the characteristics of the corresponding affer-
ent discharge initiated by the action of the stimulus energy upon
the receptor. Discrimination experiments indicate that there is not
a one-to-one parallelism between these two variables. It is held
that the number and nature of the various primary generalization
gradients are caused jointly by the nature of the stimulus energy
and the nature of the receptor response. The j.n.d. is also a joint
function of the nature of the stimulus energy and the nature of
the receptor response. It is probably because of this that general-
ization is a more simple and uniform function of distance on the
generalization continuum when the latter is measured in j.n.d.’s
than when measured in the ordinary physical units of the stimulus.

A second form of stimulus generalization applies to stimulus
compounds. The equivalence of two or more stimulus compounds
in their capacity to evoke the same reaction may depend upon (1)
the presence in each compound of certain identical (or similar)
stimulus elements or aggregates, (2) the reaction becoming condi-
tioned to the several stimulus elements or aggregates in one stim-
ulus compound, and (3) the common stimulus element in the second
stimulus compound tending to evoke the reaction much as it did in
the original compound.

The range of primary stimulus generalization has limitations,
particularly in the spread of reaction tendencies from one receptor
mode to another. Stimulus equivalence in such cases is brought
about by an indirect process known as secondary generalization.
This evolves by a series of steps: (1) energies of distinct stimulus
modes become conditioned to the same reaction by direct reinforce-
ment; (2) a second reaction may later be conditioned to one of
the stimulus energies; (3) still later, if some other stimulus also
conditioned to the first reaction but not to the second should im-
pinge on the organism, that stimulus will evoke the first reaction,
and the proprioception of this reaction will evoke the second reac-
tion. A stimulus-response chain of this kind mediates secondary

STIMULUS GENERALIZATION 199

or indirect stimulus generalization, a second form of stimulus equiv-
alence.

The summation of overlapping primary stimulus generaliza-
tions, even if definitely subliminal at their point of reinforcement
and even if each stimulus point is reinforced only once, may be
shown to raise the effective habit strength above the reaction
threshold not only at the points reinforced but at neighboring
stimuli which have never been reinforced at all. In this way are
explained both the paradox of the occurrence of superthreshold
learning where the conditioned stimulus is never exactly repeated,
and the paradox of reaction evocation where the evoking stimulus
has never been associated with the reaction evoked.

In view of the preceding considerations we may formulate
Postulate 5:

POSTULATE 6

The effective habit strength sHr is jointly (1) a negative growth
function of the strength of the habit at the point of reinforcement (S)
and (2) of the magnitude of the difference (d) on the continuum of that
stimulus between the afferent impulses of s and s in units of discrimi-
nation thresholds (j.n.d.’s) ; where d represents a qualitative difference,
the slope of the gradient of the negative growth function is steeper than
where it represents a quantitative difference.

From Postulates 4 and 5 there follows an important corollary;
because of the frequency of its use it is here given special promi-
nence:

MAJOR COROLLARY I

All effective habit tendencies to a given reaction, whether positive or
negative, which are active at a given time summate according to the
positive growth principle exactly as would the reinforcements which
would be required to produce each. *

NOTES

Mathematical Statement of Postulate 5

This postulate is expressed concisely by the equation:

sHr = aHafi-*'* (29)

where,

sH r is a a given in equation 16,
d is the difference between S and S in j.n.d.'s,

and

j' is an empirical constant of the order of .01 in the case where d is a qualita-
tive difference but of the order of .006 where d is a quantitative difference.

200

PRINCIPLES OF BEHAVIOR

Mathematical Statement of Major Corollary I

BsH R = 2i — ~~ -f- — _1_ f 1 \ n — 1 "^n* /oa\

M T Af 2 + (- l; n 1 j, (30)

wli ere Si 13 the simple sum of the items to be combined, S 2 is the sum of the prod-

ii ° f r 1 combinations taken two at a time, Z 3 is the sum of all products taken

ee a a ime, etc., and M is the physiological limit of the learning process, in
this case assumed to be 100.

The Hovland Stimulus Generalization Gradients

Hovland’s numerical values from which Figure 42 was plotted are as follows:

tr ■ . .

No. j.n.d.’s distant from,
point of reinforcement (d)

Amplitude of galvanic skin
reaction in millimeters ( A )

18.3

14.91

13.62

12.89

The negative growth function fitting these data rather well is:

A = 18.3 — 6 (1 — 10 - 0135

hirinn ^ amp ^ tu< ^ e millimeters) of the galvanic skin reaction to stimu-
lation, 12.3 represents the asymptote or limit of fall of the value of A, 6 is the

• m ° U i t ° f C ^ DgG in A due generalization, and .0135 is a constant
nn P , ,~ Qg m part on * he steepness of slope of the generalization function and in
P n he units employed. This equation is represented by the smooth curves
drawn among the data points in Figure 42.

to the eq^tion- CUrVe thr ° USh ‘ h * data poin,s of Fi K ure 43 “ rres P onds

A = 14.3 — 2.24(1 — IQ-0061 d) #

ior

The Method of Deriving from Empirical Values the Constants

Growth Function Fitted to Hovland’s Data

f„JL WaS c ° nc ' uded from a ° inspection of the data in graphic form that the

con^iH.r.KI ° bab y , Wa t a decay 0r negative growth variety with an asymptote
fom? M greater than 2ero - ie -> tha t the equation would probably be of the

A = a +(b - a ) 10-M

e Hf 0n ’ Vl the 0 ?' Ue of A *’ h » d = zero. This is given directly by
the data table as 18.3. Substituting, the equation becomes:

A = a + - q

lO^

Injthis equation, A and d are given by the table of empirical values, which leaves
two unknowns, h and a. These values are found by means of simultaneous
equations, three pairs of which may be set up from Hovland’s empirical results.

* This equation was derived by Arthur S. Day.

STIMULUS GENERALIZATION

201

Taking d = 50 for one equation, and d = 75 for the other, and substituting the
corresponding A-values from Hovland’s data, we have :

13.62 = a

18.3 - a

1Q50 A

12.89 = o -

18.3 - a

1Q76 h

Solving this pair of simultaneous equations we find that a = 11.1 and h = .008.

Setting up the two remaining significant simultaneous combinations afforded
by Hovland’s data we obtain additional values for both a and h. The whole
series of values is as follows :

25 and 50

12.9

.017

25 “ 75

12.5

.015

50 “ 75

11.1

.008

Taking the approximate central tendency of the three values for each constant
we have, a = 12.3 and h = .0135. Substituting, we have:

= 12.3 +

18.3 - 12.3

IQ 0135 d *

= 12.3 + 6 X 10“ 0135
which is the equation sought.

The Derivation of Figure 46

At the outset it is assumed that single subliminal reinforcements were made

at five points on an afferent generalization continuum: 120, 130, 140, 150 and

160 j.n.d.’s distant from a rather remote common point marked zero in the figure.

Next, the generalization value from each point of reinforcement was calcu-
lated by the equation :

bHr = 3 X 10- 0135 d (31)

For reasons given in the text (p. 186), this equation has been adapted from that
derived from Hovland's data as a sounder and more general equation for the
generalization gradient. In this way were obtained the results given in the
following table :

Difference in j.n.d. (d)

100

110

Generalized effective habit
strength (sH R ) in habs

3.0

2.57

2.20

1.88

1.61

1.18

.87

.63

.46

.34

.25

.18

.13

.10

202

PRINCIPLES OF BEHAVIOR

From this table were obtained the generalization values sloping downward
and away in each direction from the crests which appear in the lower part of the
figure at the five points of reinforcement.

The next step in the process was to combine the overlapping generalization
tendencies which appear at every point on the stimulus continuum as five items.
For example, at 120 the first item is 3.0 because here d = zero. Next, there is
the generalization value of the gradient originating at 130, which is 10 j.n.d.’s
distant. By the above table this would yield 2.20 habs. Then there is the
generalization value originating at 140. Since 140 is 20 j.n.d.’s from 120, d = 20,
which, by the above table, corresponds to a generalization value of 1.61 habs.
In a similar way the other two 8 H R values with d’s of 30 and 40 are shown by
the table to be 1.18 and .87, respectively.

We thus have the problem of combining 3.0, 2.2, 1.61, 1.18, and .87 habs.
How shall this be done? The hypothesis which fits in best with various related
empirical observations is that the generalized effects of learning summate in
the same way as do the effects of repetitions in the learning process. The prin-
ciples according to which repetitions of reinforcement combine to produce habit
tendencies have been explained in considerable detail in Chapter VIII and are
stated concisely in Postulate 4 and in equations 1, 26, and 30. It will be sufficient
here only to say that each repetition was supposed to increase the amount of
habit strength by a constant factor (e.g., 1/10) of the learning potentiality not
yet realized in actuality. This means that the amount of habit strength con-
tributed by one repetition of reinforcement late in the learning process is very
much less than that contributed by one repetition at the beginning. In a similar
manner, a block of five repetitions given late in the learning process will contribute
less to the habit strength than an exactly similar block of five repetitions early
in the process. It was assumed that the five generalized tendencies summate
just as would the number of repetitions required to produce each. Thus one
might substitute the value of bH r in equation 26 :

bH r = m(l - 10-^0

and solve for N in the case of each of the five values listed above, add together
the set of N’s thereby obtained, and, finally, substitute the sum of the five N’s
in the same equation, this time solving for sH R \ the result of the last operation
would constitute the summation required.

The above procedure, while conceptually simple, is very clumsy mathe-
matically. Accordingly the mathematical implications of the assumption have
been worked out for the general case where n values must be eliminated. This
takes the form of the equation given in the second terminal note above! (30).
The meaning of that equation will be made clear by the following example":

Zi = 3.00 + 2.20 + 1.61 + 1.18 + .86 = 8.85.

2a = 3.00 X 2.20 + 3.00 X 1.61 + 3.00 X 1.18 + 3.00 X .86 + 2.20 X 1.61

+ 2.20 X 1.18 -f 2.20 X .86 + 1.61 X 1.18 -f- 1.61 X .86 + 1.18 X .86

= 6.60 + 4.83 + 3.54 + 2.58 + 3.54 + 2.60 -f 1.89 + 1.90 1.38 4- 1 01

= 29.87

Z 3 = 3.00 X 2.20 X 1.61 + 3.00 X 2.20 X 1.18 + 3.00 X 2.20 X .86 + 3 00
X1.61 X 1.18 + 3.00 X 1.61 X .86 + 3.00 X 1.18 X .86 + 2.20 X 1.61
X1.18 + 2.20 X 1.61 X .86 + 2.20 X 1.18 X .86 + 1.61 X 1.18 X .86
= 10.63 + 7.79 + 5.68 + 5.70 + 4.15 + 3.04 + 4.18 -f 3.04 + 2.24 + 1.63
= 48.08

STIMULUS GENERALIZATION

203

ZU = 3.00 X 2.20 X 1.61 X 1.18 + 3.00 X 2.20 X 1.61 X .86 + 3.00 X 2.20
X 1.18 X .86 + 3.00 X 1.61 X 1.18 X .86 + 2.20 X 1.61 X 1.18 X .86
= 12.54 + 9.14 + 6.70 + 4.90 + 3.59
= 36.87

2* = 3.00 X 2.20 X 1.61 X 1.18 X .86
= 10.78.

ssHr = 8.85 —

= 8.85 -
= 8.55

29.87 . 48.08

36.87

10.78

100 1 10,000 1,000,000 100,000,000

.30 + .0048 - .000037 + .0000001 1

which is the value of bsHr at 120 on the upper or summation graph in Figure 46.
All of the other points in the summation curve were computed in an analogous
manner.

In general this method of summation yields a value appreciably less than
would be obtained by the simple addition of the items summated, the shrinkage
being greater the nearer the individual items approach the magnitude of the
physiological limit (M), in this case, 100. Because of the relatively s ma l l size
of the items, the shrinkage here is alight.

REFERENCES

1. Birge, J. S. The role of verbal responses in transfer. PhD. thesis, Yale

University, 1941.

2. Guthrie, E. R. Conditioning as a principle of learning. Psychol. Rev.,

1930, 37, 412-428.

3. Hilqard, E. R., and Marquis, D. G. Conditioning and learning. New

York: D. Appleton-Century Co., Inc., 1940.

4. Hovland, C. I. The generalization of conditioned responses: I. The sen-

sory generalization of conditioned responses with varying frequencies
of tone. J. Oen. Psychol., 1937, 17, 125-148.

5. Hovland, C. I. The generalization of conditioned responses: II. The sen-

sory generalization of conditioned responses with varying intensities of
tone. J. Genet. Psychol., 1937, 51, 279-291.

6. Hull, C. L. Quantitative aspects of the evolution of concepts. Psychol.

Monogr ^ 1920, 28, No. 123.

7. Hull, C. L. Learning: H. The factor of the conditioned reflex. Chap-

ter 9 in A handbook of general experimental psychology, C. Murchison,
editor. Worcester, Mass.: Clark Univ. Press, 1934.

8. Hull, C. L. The problem of stimulus equivalence in behavior theory.

Psychol. Rev^ 1939, Jfi, 9-30.

9. Humphreys, L. G. Generalization as a function of method of reinforce-

ment. J. Exp. Psychol., 1939, 25, 361-372.

10. Lashley, K. S. The mechanism of vision: XV. Preliminary studies of

the rat’s capacity for detail vision. J. Gen. Psychol ^ 1938, 18, 123-
193.

11. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

12. Shipley, W. C. Indirect conditioning. J. Gen. Psychol ., 1935, 12, 337-

357.

13. Thorndike, E. L. Educational psychology, n. The psychology of learn-

ing. New York: Teachers College, Columbia Univ., 1913.

CHAPTER XIII

Some Functional Dynamics of Compound Conditioned

Stimuli

Most accounts of conditioning experiments tend enormously to
minimize the actual complexity of the factors involved. Indeed,
this is almost a necessity; if the reports of such experiments should
contain a really complete description of the process the reader
would be so swamped in detail that he might easily fail to under-
stand the main point of the experiment. The same expository diffi-
culty is encountered by behavior theorists in perhaps an even more
aggravated form and has led to the same type of misrepresentation.
An example of this is our own use of the symbol 8 Hr • There is
small doubt that such expository over-simplifications in the ac-
counts of learning and other behavior situations have genuinely
misled many persons beginning the study of behavior and, pos-
sibly, in some cases even the investigators themselves; they cer-
tainly have produced much misapprehension among the philosophi-
cal critics of behavior theory, who as a rule have little or no first-
hand knowledge of the phenomena concerned and so are especially
prone to such misunderstandings. In Chapter XII something was
done to remedy this wholly natural yet regrettable situation by
considerably expanding the concept of habit with respect to the
stimulus; a parallel expansion on the response side will be pre-
sented in Chapter XVII. In the present chapter we shall seek to
clarify the concept of the stimulus (S) still further by deriving a
number of elementary corollaries from the conditions under which
learning occurs when considered in conjunction with certain
primary principles; the latter are for the most part already familiar
to the reader.

THE COMPLEXITY OF THE “STIMULUS” OF A TYPICAL

CONDITIONING SITUATION

In order to clarify to some degree the actual complexity of the
“stimulus” involved in typical habit formation, let us consider in
a little detail this aspect of what is usually regarded as one of the
more simple learning situations, that of the Pavlovian conditioned

204

COMPOUND CONDITIONED STIMULI 205

reflex (tf). Previous to the beginning of the conditioning process
the dog has had no food for 24 hours. It stands upon a table in
the laboratory, being kept in place by loose bands attached to a
wooden framework. Some weeks previous to the experiment one
of its salivary ducts has been surgically diverted so that saliva is
discharged through a fistula in the side of the animal’s face. When
the experiment actually begins, a capsule is cemented over the
fistula in such a way that it collects the saliva seeping through the
opening; the pressure in the capsule resulting from the entrance of
the saliva is transmitted by a rubber tube to a sensitive register-
ing device.

An electric buzzer is sounded near the dog for five seconds;
two seconds after the termination of the buzzer action a small
quantity of meat powder is blown into the animal’s mouth by means
of a rubber tube held in place by a kind of muzzle. This powder
is eaten, with a profuse accompanying flow of saliva. After a few
repetitions of this sequence, the dog begins secreting saliva during
the seven-second interval between the beginning of the sound and
the delivery of the meat powder, showing that the conditioned
reflex has been set up.

The above summary description of the conditioned reflex experi-
ment is a fairly typical example of the accounts usually given; the
only conditioned stimulus element specifically mentioned as active
in the situation is the buzzer. As a matter of fact, the buzzer
vibration makes up only a small part of the total number of stim-
ulus components involved. Moreover, the wave pattern of the buz-
zer itself, as revealed by the cathode ray oscillograph, is an exceed-
ingly complex phenomenon and doubtless stimulates simultaneously
a very large number of the ultimate auditory receptors in the
cochlea.

Among the many additional components of the conditioned
stimulus (5) not ordinarily mentioned are: the fact that the ani-
mal’s two ears receive the buzzer vibrations with different intensity
or in different phase, depending on (1) the direction of the bell
from the dog’s head and (2) the orientation of the head at the
moment; the pressure of the dog’s feet against the table top upon
which it stands; the pressure of each of the three or four restrain-
ing bands upon the skin receptors of the dog’s neck, thighs, etc.:
the biting of a number of insects which may be hidden in the
dog’s hair; the contact of the capsule over the fistula; the pressure
of the muzzle against the dog’s head; the pressure of the rubber

20 6

PRINCIPLES OF BEHAVIOR

tube in the dog’s mouth; the odor of the rubber from which the
tube is made, together with a large number of miscellaneous odors
to which the human olfactory receptors may or may not respond;
the multitude of visual stimuli of light, shade, spatial combinations,
etc., arising from the laboratory lamps and reflected from millions
of points within the dog’s visual field; the proprioceptive stimu-
lation arising from the external and internal muscles of the dog’s
eyes as they fixate one object after another about the laboratory;
the infinite number and variety of proprioceptive impulses origi-
nating in the several parts of the other muscles of the animal’s
body as they are employed in the maintenance of the postures
taken from moment to moment; the too-little understood stimula-
tions associated with the bodily state resulting from food, water,
and sexual privation, rectum and bladder pressure, etc.; and, finally,
the perseverative traces of all the multitude of stimuli recently
acting, whether the stimulus energy is continuing to act at the
moment or not. The conditioned stimulus in the experiment under
consideration includes all of the immensely complicated stimulus
elements here enumerated and many more besides; nevertheless
this list, incomplete as it is, should aid the reader somewhat in
overcoming the misleading suggestion of singularity and simplicity
otherwise likely to be conveyed by the S of the symbol, gH R .

THE DISTRIBUTION OF HABIT STRENGTH ACQUIRED BY THE
SEVERAL COMPONENTS OF A STIMULUS COMPOUND

The law of primary reinforcement as formulated in Chapter VI
presented, in the interest of introductory expository clarity, the
ultra-simple view of the conditioned stimulus which we have just
been at some pains to rectify. We must now consider the opera-
tion of this principle under the present expanded conception, par-
ticularly as it applies to the several types of components which
may be found in a stimulus compound.

According to the “law of reinforcement” laid down earlier (p.
80), every one of the receptor discharges and receptor-discharge
perseverations active at the time that the to-be-conditioned reac-
tion occurs must acquire an increment of habit strength (A a H R ).
The enunciation of this principle, coupled with the recognition of
the multiplicity and variety of these afferent elements, at once
raises numerous critical questions, one of which is: Are these incre-

COMPOUND CONDITIONED STIMULI

207

ments of habit strength all of the same magnitude? The answer
to this question is quite definitely that the increments of habit
strength acquired by the several afferent discharges arising from
the various stimulus aggregates represented by such words as
“buzzer sound/’ “odor of food,” “sight of food cup,” “pressure of
restraining bands,” etc., differ widely. In this respect the situation
is believed to be substantially as represented in Figure 47, where
the thickness of the broken lines connecting the several stimulus

aggregates (S’s) represents the varying mag-
nitudes of the increments of habit strength
acquired by them at a given reinforcement.
The S’s shown in the diagram are, of course,
far too few to more than suggest the number
of actual stimulus elements, or even the num-
ber of the aggregates of stimulus elements, 1
in the typical conditioning situation.

Recognition of the variability in the incre-
ments of habit strength acquired by the sev-
eral stimulus components of a conditioning
situation at once raises the question of the
principles according to which the differential

Fig. 47. This figure
represents the sheaf
of habit tendencies

(As//*) presumably-
set up by each rein-
forcement. The thick-

magnitudes of habit-strength loadings arise.
This question does not permit a very definite
answer, though a certain amount of experi-
mental effort has been directed to this end.

ness of the dashes
leading to the several
arrow points is in-
tended roughly to rep-
resent the differences

One bit of evidence comes from the Kappauf-
Schlosburg experiment described above (p.
166), which showed that stimuli which have
continued to act on a receptor without change
for some time have a greatly diminished ca-
pacity for acquiring habit loadings. Partly
for this reason it is probable that static, i.e.,

in the magnitude of
the habit increments
or habit loadings con-
necting the several
stimulus elements in
the conditioning situ-
ation, with the reac-
tion process.

unchanging, elements or aggregates in a conditioned stimulus

situation are considerably less potent in the acquisition of habit-
strength loadings than are the more dynamic, i.e., changing, ele-
ments or aggregates. This probably is why investigators so fre-
quently neglect to take into consideration the static or constant

1 A stimulus element is defined as the action of a stimulus energy upon a
single receptor organ, such as a single rod of the retina or a single touch
organ of the skin. A stimulus aggregate is a group of stimuli which ordi-
narily begin and end concurrently and, in general, combine to perform the
same adaptive functions.

208 PRINCIPLES OF BEHAVIOR

elements of conditioned stimulus situations, sometimes with unfor-
tunate results.

A second factor which may be of some importance in deter-
mining the habit-strength loading acquired by a stimulus aggre-
gate in a learning situation is the intensity of the stimulus energy.
Pavlov reports ( 6 , p. 142), on the basis of admittedly inadequate
empirical data, that when two stimulus energies of different inten-
sities operate on the same receptor simultaneously, e.g., two differ-
ent tones, the stronger stimulus receives a greater increment of
habit strength.

A third factor is the receptor or “analyzer,” which receives the
stimulus energy. Pavlov reports ( 6 , p. 143), again without ade-
quate supporting evidence, that, other things equal, tactual stimuli
seem to acquire stronger habit loadings than do thermal stimuli
and that auditory stimuli are stronger in this respect than visual
stimuli. Recently Zener has reported orally a well-controlled
study which fully establishes this proposition for relatively low
stimulus intensities.

A fourth possible factor in determining the relative habit load-
ings of the several components of a stimulus compound concerns
whether or not a stimulus appears in a large number of conditioning
situations which require a wide variety of reactions to bring about
need reduction, and also in many situations requiring no reaction
whatever. For example, daylight is present as a visual stimulus
component in thousands of different reaction situations. It seems
likely that the great number of reactions to which this stimulus
component is conditioned early in life, coupled with incidental
extinction effects which necessarily result from such a state of
affairs, would soon largely blur out the capacity of such stimuli
to be conditioned to any reaction in particular.

A fifth factor which follows as a kind of corollary to the factor
of intensity is that a stimulus component which has previously
been conditioned to a reaction involving strong autonomic or
emotional aspects, e.g., a fear reaction, will presumably acquire
in this indirect way a stronger habit loading than would a com-
ponent not so conditioned. This would be expected on the assump-
tion that the proprioceptive and other receptor discharges entailed
by the occurrence of the conditioned reaction in question would
constitute a relatively intense stimulus which, as such, would ac-
quire a correspondingly heavy habit loading from the reinforce-
ment process. A very weak stimulus through lack of vigorous

COMPOUND CONDITIONED STIMULI

209

competition or through especially powerful reinforcement in an
earlier conditioning situation might thus acquire control of a reac-
tion yielding a powerful stimulus and in this indirect way attain
the appearance, in subsequent conditioning situations, of having
itself a strong capacity for acquiring habit loadings. It seems
likely that this mechanism explains to a considerable extent the
role in the learning process of what is reported introspectively as
“attention.”

Unfortunately as yet very little experimental effort has been
directed to the solution of these problems. For this reason most
of the suggestions listed above must be regarded as hardly more
than conjectures suitable as points of departure for future inves-
tigations.

THE JOINT REACTION-EVOCATION POWER OF COMPOUND

STIMULUS AGGREGATES

Among the problems precipitated by the highly compound char-
acter of the conditioned stimulus there is the question of the
relative action-evoking power (and, presumably, of effective habit
strength) of a stimulus element or aggregate ( S ) when acting in a
stimulus compound as contrasted with that when it is acting alone.
There are two cases; one is that in which the stimulus components
are separately conditioned to the same reaction and are later tested
for power of reaction evocation by being presented to the subject
simultaneously as a stimulus compound. The other is that in
which the stimulus components are presented simultaneously as a
compound during the conditioning process and are later tested for
power of reaction evocation by being presented to the subject sepa-
rately. Here, as in so many other aspects of behavior science,
really adequate empirical investigations are largely lacking. How-
ever, some experimental results are available which are useful for
illustrating the nature of the problems and for suggesting plausible
hypotheses, even if not adequate to serve as a basis for final deci-
sions.

We proceed now to the consideration of the empirical evidence
concerning the case where the stimulus components are conditioned
separately to the same reaction. In one illustrative experiment
eight human subjects were first conditioned to a weak light, the
unconditioned stimulus being a brief electric shock and the response
recorded, the galvanic skin reaction (2). Next, a weak vibratory

210

PRINCIPLES OF BEHAVIOR

stimulus applied to the forearm was paired with the shock and
thus conditioned to the galvanic skin reaction. When presented
alone the light (Si) evoked a mean reaction (#i) of 3.5 millimeters,
and the vibrator ( S v ) evoked a mean reaction ( R v ) of 3.6 milli-
meters. When both stimuli were presented simultaneously (Si +V )>
they jointly evoked a mean reaction (#! + *) of 4-4 millimeters.

Since the two stimulus components were of almost equal reac-
tion-evocation strength when presented separately, it is reasonable
to suppose that they contributed about equally to the joint evoca-
tion when they were presented simultaneously, i.e., that each con-
tributed approximately half of the 4.4 millimeters of the joint
evocation, or 2.2 millimeters. These results, which are typical of
a number of fairly comparable sets now available, indicate a defi-
nite shrinkage in the reaction-evocation power (and so, presumably,
of effective habit strength) at the command of stimulus compo-
nents in combination, as compared with that when acting sepa-
rately.

A simple quantitative index of this shrinkage is obtained by
dividing the amplitude of the reaction evoked by the joint stimu-
lation by the sum of the amplitudes evoked by the separate stimu-
lations:

^ + * _ 4.4 _ 4v4 _

Ri + R, 3.5 + 3.6 7.1 “ # -

This is interpreted to mean that a stimulus when presented jointly
with another evokes only about 62 per cent as great a reaction as
it does when presented separately, thus showing a shrinkage of
38 per cent, or about a third. A comparable experiment (£), also
employing eight human subjects, yielded the following values:

R* + c _ 3.91 _ 3.91

Ri + R, 2.2 + 3.7 5.9 "

.66 +.

This experiment also shows a shrinkage of almost exactly one-third.

The second case of the reaction-evocation power of compound
stimuli is the reverse of that just considered. This is illustrated
by a simple permutation of the series of experiments outlined
above, the present experiment likewise employing eight human sub-
jects (£). The light and the vibrator, presented simultaneously,
were paired with the shock. The mean amplitudes of the condi-
tioned galvanic skin reaction evocable by the compound and by
the separate components were then determined. It was found that
the light alone (Si) evoked a mean reaction amplitude of 2.5 milli-

COMPOUND CONDITIONED STIMULI

21 1

meters; the vibrator alone (S„) evoked a mean reaction amplitude
of 2.9 millimeters; and the two components presented as a com-
pound (jSi + v ) evoked a mean reaction amplitude of 3.3 millimeters.
Despite the difference in the arrangement of this experiment, the
shrinkage of the reaction-evocation power of the stimulus com-
ponents when in a compound as compared with that when pre-
sented separately may be calculated by the same formula:

R l + . 3.3 _ 3.3

R t + R , 2.5 + 2.9 5.4

.61 +.

The calculation yields a value within the range of experimental
error of that obtained from the experiments falling under case one.

Another exactly comparable experiment from the same study,
also employing eight human subjects, yielded the following result:

R 1 + 9 _ 2.8 2.8 _ eg .

Ri + R. 1.4 + 2.8 4.2 * DO ~ Vm

To the two cases of compound stimulation already considered
there may now be added a third case which is of special interest
because it is neutral so far as possible interaction effects arising
from the circumstances surrounding the conditioning are concerned.
This is illustrated by an experiment employing eight human sub-
jects (£) in which the S > R connection was the result of what is

called sensitization , i.e., it was set up by merely administering the
shock without the latter being paired with either the light or the
vibrator. This experiment yielded the following result:

+ » 3.2 _ 3.2 _ _

R, +R, 2.2 + 2.9 5.1 ’

which is in close agreement with the results of the other four
experiments.

Thus, from the point of view of the quantitative index, ! .

Ri Rv

all three cases, so far as the available evidence goes, agree in
showing that where two stimulus aggregates are concerned a shrink-
age of about one-third in reaction-evocation power occurs; i.e.,
from this particular point of view no marked difference between
the three cases appears. This may very well be the situation
which will finally be revealed by further experiment.

There are, however, certain indications reported by Pavlov ( 6 ,
p. 141) to the effect that where two dynamic stimulus components

2 I 2

PRINCIPLES OF BEHAVIOR

are jointly conditioned to a reaction, the stronger of the two may
completely dominate, obscure, or “overshadow” the weaker com-
ponent in that when the latter is presented separately it will evoke
no reaction whatever, though in certain indirect ways it may be
shown to possess some kind of functional connection with the reac-
tion. In this context it will be recalled that in the fourth example
given above, which involved a reaction conditioned to a compound
stimulus, the mean amplitude of reaction evoked by the compound
(2.8 millimeters) was exactly the same as that evoked by the
stronger of the two stimulus components, the cutaneous vibrator;
the addition of the light to the combination seems to have added
nothing to the mean amplitude of reaction. This outcome might
quite possibly have been due to experimental “error,” i.e., to the
fact that the sample of data collected was too small to yield a
sufficiently precise indication of the relative influence of the several
factors involved.

After weighing the various bits of experimental evidence avail-
able, the most plausible empirical generalization concerning the
dynamics of compounds of conditioned stimuli in the reaction evo-
cation is that stimulus aggregates conditioned to the same reaction
possess, irrespective of whether the stimulus aggregates were con -
ditioned to the reaction as separate entities or as a stimulus com-
pound, (1) a smaller power of reaction evocation when presented
jointly than when presented separately, but ( 2 ) a larger joint power
of reaction evocation than does any single component when the
latter is presented separately.

THE PRINCIPLES OF HABIT SUMMATION AND OF MONOTONIC
HABIT-REACTION RELATIONSHIP AS APPLIED

TO STIMULUS COMPOUNDS

With some examples of the coarse empirical reaction-evocation
dynamics of stimulus compounds before us, the question arises as
to whether the above empirical generalizations represent primary
principles or secondary principles derivable from a combination of
primary principles already in the system. The answer to this type
of question depends on whether or not the principle can be derived
from these other and supposedly more elementary principles; if so,
it is a secondary principle; if not, it may be a primary principle.
It should be said at once that by this test, the dynamics of reaction
generalization are second-order, or derived, principles. The major

COMPOUND CONDITIONED STIMULI

2*3

portion of these phenomena appear to be mainly dependent upon,
i.e., derivable from, two principles already more or less familiar.

The first of the primary principles in question concerns the
manner in which two or more homogeneous habits, i.e., those in-
volving the same reaction, summate to produce a joint habit
strength. We encountered a situation of this kind above (p. 195)
when we considered the summation of habit strengths arising from
the overlapping of stimulus generalization gradients both of which
were conditioned to the same reaction. In that case we had the
physiological summation of habit tendencies generated by the con-
ditioning of the same reaction to two distinct stimuli but, through
the process of generalization, brought to bear on the evocation of
the reaction by the impact of a single stimulus aggregate. The
present situation presumably has the same dynamic state of affairs
brought about by the simultaneous action of two stimulus aggre-
gates each of which has a tendency independently to evoke the
same reaction. As in the case of overlapping stimulus generaliza-
tions (Chapter XII, p. 199), the quantitative principle of summa-
tion is given by Major Corollary I. As applied specifically to the
present situation, this may be restated as follows: If two or more
stimulus aggregates, each independently conditioned to the same
reaction, impinge simultaneously on the receptors of the organism
in question, the effective habit strengths borne by the several stim-
ulus aggregates summate to produce a joint habit strength as would
the separate effects of the number of reinforcements necessary to
produce each, if such reinforcements were to be given consecutively
in some standard reinforcement sequence.

Thus, suppose that one stimulus aggregate, such as a weak
light, has a habit strength to the evocation of a given reaction of
34.39 habs, and a second stimulus aggregate, such as a cutaneous
vibrator, has by independent reinforcement a habit strength of
46.86 habs to the evocation of the same reaction. Now, by Table 1
(Chapter VIII, p. 115), 34.39 habs would (under certain assumed
conditions) be produced by four reinforcements, and 46.86 habs
would be produced by six reinforcements. On the above principle
it follows that the physiological summation of the two habit
strengths would yield a joint habit strength equal to that which
would be produced by 4 -f- 6 = 10 reinforcements, which, by Table
1, equals 65.13 habs.

The second principle presumably operating here is that of the
monotonic habit-reaction relationship (p. 326 ff.). This is that

2I 4

PRINCIPLES OF BEHAVIOR

the strength of a reaction tendency is an increasing linear function
of the effective habit strength mediating it; i.e., if

bMb > sJIr,

then

Rsi > Rsf

SOME COROLLARIES OF THE PRINCIPLES OF HABIT-STRENGTH
COMBINATION AND OF THE MONOTONIC HABIT-REAC-
TION RELATIONSHIP APPLIED TO CONDITIONED

STIMULUS COMPOUNDS

It follows from the above principles that the sum of the reaction
amplitude mediated by 34.39 habs and of that mediated by 46.86
habs will be an increasing function of 34.39 -f- 46.86, or 81.25,
whereas the magnitude of the reaction mediated by the joint action
of the two stimuli will be a corresponding function of 65.13. How-
ever,

65.13 < 81.25.

We accordingly formulate Corollary I as follows:

I. The amplitude x of the reaction evoked by two stimulus ag-
gregates acting jointly will be less than will be the sum of the
reaction magnitudes evoked by the respective stimulus aggregates
acting separately. Stated in a formal manner, Corollary I becomes
the following inequality:

Ri + • < Ri + R»-

Returning, now, to our empirical data we find ample illustra-
tion in all five experiments. For example, in the first experiment
we find,

Ri+ 9 = 4.4

and

Ri + R, = 3.5 + 3.6 = 7.1

i.e.,

4.4 < 7.1,

which fully satisfies the above inequality.

A secondary corollary which flows from the same assumptions
concerns the reaction magnitude mediated by the summation of

1 This statement concerning amplitude appears to hold only for certain
reactions such as the galvanic skin reaction, here used as illustrative mate-
rial, the salivary reaction, and the lid reaction. For certain other types of
reaction such as are normally produced by the striated muscles, probability
of reaction evocation (p) at stimulation, latency or resistance to experimental
extinction will need to be substituted for amplitude.

COMPOUND CONDITIONED STIMULI

2I 5

two habits as contrasted with that mediated by either habit alone.
We have seen that the joint strength of a habit of 34.39 habs and
one of 46.86 habs amounts, theoretically, to 65.13 habs, which is
greater than either habit strength taken alone, i.e.,

34.39 < 46.86 < 65.13.

Therefore,

Rs t < Rsi+Sf

Generalizing from the above we arrive at our second corollary:

II. The magnitude of the reaction tendency evoked by any one
of a number of stimulus aggregates conditioned to the same reac-
tion will be less than that evoked by two or more of them acting
simultaneously. Empirical illustration of this corollary is seen in
four of the five experiments cited above. Thus 4.4 is greater than
either 3.5 or 3.6, and 3.9 is greater than either 2.2 or 3.7. The
slightly discordant results of the fourth experiment are probably
due to limitations in the size of the sample.

Thirdly, suppose that three equally potent stimulus aggregates
possess a joint strength of 71.76 habs. Now, 71.76 habs corresponds
(Table 1) to twelve standard reinforcements. Thus the three stim-
ulus components must have independent strengths equivalent to
one-third of 12, or four standard reinforcements, each of which, by
Table 1, corresponds to 34.39 habs. According to the principle of
habit strength summation, two of these, if taken together, would
summate to a habit strength equivalent to eight standard rein-
forcements, which (Table 1) would yield 56.95 habs. But,

34.39 < 56.95 < 71.76

i.e.,

sJIr < s\Hr + sJJr < s\Hr + s*Hr “ I - siH r .

It follows from this and the principle of the monotonic habit-reac-
tion relationship that,

R Si < R Si + St < Rs t + St + Si*

Generalizing, we formulate our third corollary:

III. If a number of stimulus aggregates, all equally conditioned
to the same reaction, impinge upon the organism simultaneously,
the larger the number of such stimulus aggregates active on a given
occasion, the greater will be the amplitude of the evoked reaction.

21 6

PRINCIPLES OF BEHAVIOR

THE PRINCIPLES OF AFFERENT INTERACTION AND PRIMARY

STIMULUS GENERALIZATION

The third principle upon which the dynamics of reaction evoca-
tion by stimulus compounds depends is that of afferent neural
interaction (Postulate 2, p. 47). In the present context that
principle may be stated as follows (3, p. 77) : Concurrent afferent
impulses (s) arising from the impact of distinct stimulus energies
(S) on receptors are appreciably modified by each other before
they reach that portion of the central nervous system where they
initiate the efferent impulses (r) which ultimately evoke reac-
tions ( R ).

The relevance of the afferent interaction hypothesis to the
dynamics of stimulus compounds arises from the fact that two
stimulus aggregates may, as we have already seen, act either inde-
pendently of each other or in combination. But, by the principle
of afferent interaction, the afferent impulses which originate in a
given stimulus aggregate are somewhat different on reaching the
more central portions of the nervous system when they occur con-
currently with the afferent impulse arising from the action from
some other stimulus energy than when the second afferent impulse
is not occurring. Thus, two stimulus energies S t and S g when act-
ing separately initiate afferent impulses Sj and s g , but if acting at
the same time these afferent impulses interact, changing each other
to some extent so that by the time they reach the more central

portions of the nervous system s, has changed to s lt and s. has
changed to s*.

Now, suppose that Sj and S g have each been conditioned sepa-
rately to reaction, R. This process would produce the relationship,

51 —> Si > r —> R

and

5 2 — > s 2 >r—>R.

However, when S t and S g act simultaneously as a stimulus com-
pound, the afferent impulse which tends to r > R in the one case

is not Sj but s lt and in the other case it is not s g but s g .

At this point the principle of afferent interaction in stimulus
compounds is supplemented by the principle of the primary stim-
ulus generalization gradient (Postulate 4, p. 178). In the present
context this may be restated as follows: The habit strength at the
command of an afferent impulse is a decreasing growth function

COMPOUND CONDITIONED STIMULI

217

of the difference ( d ) between the evoking afferent impulse (s) and
the afferent impulse ($) originally conditioned to the reaction. Thus
the effective habit strength commanded by S, will be less than that
commanded by s, as originally conditioned, i.e.,

saHr < s\Hr'

From the two last mentioned principles there may be deduced
a number of corollaries, some of which limit the generality of the
corollaries just derived, and vice versa. Actually, all the principles
here under discussion are operating in every hypothetical situation
considered in the present chapter; in the interest of expository
clarity their action is here taken up separately. They will be con-
sidered jointly in Chapter XIX (p. 349 ff.).

SOME COROLLARIES OF THE PRINCIPLE OF AFFERENT INTERACTION
AND OF PRIMARY STIMULUS GENERALIZATION APPLIED

TO STIMULUS COMPOUNDS

From the inequality last considered and the monotonic habit-
reaction relationship it follows that,

< R*.

Generalizing, we arrive immediately at our fourth corollary:

IV. If a stimulus aggregate has been conditioned to a reaction,
and if, later , this stimulus aggregate is presented to the subject in
conjunction with an alien stimulus aggregate not hitherto condi-
tioned to the reaction in question, the strength of the reaction ten-
dency evoked by the stimulus combination will be less than will
that evoked by the stimulus aggregate as originally conditioned.

This corollary is exemplified in one form of what Pavlov called
external inhibition, i.e., the form where the “extra” stimulus which
produces the external inhibition does not itself evoke a reaction
conflicting with that normally evoked by the conditioned stimulus.
In this connection Pavlov remarks ( 6 , p. 45) :

If one experimenter had worked with a dog and established some firm
and stable conditioned reflex, conducting numerous experiments with them,
when he handed the animal over to another experimenter to work with,
all the reflexes disappeared for a considerable time. The same thing
happened when the dog was changed over from one research room to
another.

The interpretation is that the change in the stimulus situation
produced by the presence of a new experimenter or by a different

2 I 8

PRINCIPLES OF BEHAVIOR

room so modified the afferent impulses arising from the original
conditioned stimulus that its reaction evocation power sank below

the reaction threshold {3, p. 78).

What appears to be the same thing but in an even clearer form
has been reported by Lashley ( 5 , pp. 140-141). Rats were trained
to jump to a black card bearing a white triangle and not to jump
to a similar card bearing a white cross of the same area. After
the animals had attained more than 95 per cent of correct choices,
new cards were substituted which contained the same figures, with
four small figures added. The score of “correct” choices with the
new cards fell to 90 per cent. The interpretation is that the four
additional figures so changed the afferent impulses arising from the
triangle and the cross that the effective habit strength at the com-
mand of each was appreciably weakened, which naturally decreased
the accuracy of the discrimination. It would appear that this sec-
ondary principle (external inhibition) is operating on a very large
scale in all the higher organisms including ourselves, with whom it
is often called “distraction of the attention.”

A variant of the situation just considered is that in which two
stimuli have been conditioned to the same reaction separately and
then are presented to the organism simultaneously. The interaction
of the two afferent impulses upon each other when occurring con-
currently may be expected to reduce the effective habit strength of
each so that the amplitude of the reaction evoked by them jointly
will be appreciably less than would result from the simple summa-
tion of the two habit strengths. Thus suppose that St and S t have
each been conditioned separately to R to the extent of 70 habs,
and that the afferent interaction effect of each on the other is
of an extent sufficient to change the effective habit strength of
'SjHr and 5 *H R from 70 to 40 habs each. Now, the physiological
summation of two habit strengths of 70 habs each yields 91 habs,
whereas the summation of two habit strengths of 40 habs each
yields 64 habs. Accordingly, the summation mechanism alone
would reduce the effective habit strength of the two stimuli to
91 habs, and the interaction mechanism would reduce it still fur-
ther — from 91 habs to 64 habs, which is less than the original value
of the habit strength commanded by each individual component.
It is evident that if the interaction effects are sufficiently great they
may completely neutralize any summative effects otherwise re-
sulting from combining homogeneous conditioned stimuli, and even
yield a weaker reaction tendency than that evoked by either stim-

COMPOUND CONDITIONED STIMULI 219

ulus aggregate acting alone. Our fifth corollary accordingly is as
follows:

V. If two stimuli separately and equally conditioned to the same
reaction are later presented simultaneously , and their afferent neural
interaction effects are sufficiently great, they will jointly evoke a
weaker reaction tendency than would be mediated by either one
of the stimuli acting alone. Corollary V, it will be observed, con-
stitutes a limitation on the generality of Corollaries I and II, the
extent of the limitation depending on the magnitude of the afferent
interaction involved in the change of stimulus conditions.

An experimental example of some historical interest illustrates
this point. Shepard and Fogelsanger (7) had subjects learn paired

nonsense syllables in the arrangement A > C and B > C.

Later, syllable A and syllable B were presented together. It was
found that instead of decreasing the reaction latency of the evo-
cation of syllable C (which would result from the physiological
summation of two habit strengths), the joint presentation actually
increased the weighted mean latency from 1135 milliseconds to
2538 milliseconds, the two latencies yielding a ratio of 1 to 2.323.

In the light of Corollary V these results present no paradox
whatever. However, in some quarters they have produced a certain
amount of confusion in the interpretation of rote-learning phe-
nomena. For example, Kohler (4, p. 316) remarks of this particular
experiment,

If it were not for organization one should expect that, both excitants
working in the same direction, the syllable associated with them would
be more easily reproduced than a syllable for which there was only one
excitant. But the contrary was observed; it seemed as though some
inhibition were in the way of reproduction when it was aroused by two
excitants.

According to the present view, no inhibition whatever is involved
in the determination of the Shepard-Fogelsanger results, and the
unspecified organization or configurational factor is in reality a
combination of afferent neural interaction and stimulus general-
ization.

A sixth corollary arises from a reversal of the situation pre-
sented by Corollary V. Suppose that two equally potent stimulus
aggregates, 5, and S e , have been conditioned jointly to the same
reaction to the extent of 65.1 habs, and that later S t is presented
to the subject separately. Since (Table 1) 65.1 habs corresponds
to ten standard reinforcements, simple habit summation dynamics

220

PRINCIPLES OF BEHAVIOR

would give S x the equivalent of five reinforcements, or 40.95 habs.
Now, the interaction between afferent impulses arising from St and
S g will change what otherwise would have been s x to St. Since
the reaction is actually conditioned to (rather than to Sj), it fol-
lows that later, when Sj is presented separately from Sg and other
dynamic stimuli, the afferent impulse thus sent into the central
nervous system will be s lt rather than s t . But by the stimulus
generalization gradient, Sj will command a weaker effective habit
strength than will s it and so will evoke a weaker reaction tendency
than is proper for a habit strength of 40.95 habs, depending upon
the amount of afferent interaction which occurred in the original
conditioning situation. If the interaction effects are of sufficient
intensity, the effective habit strength at the command of one of
the stimulus aggregates under the assumed conditions may not be
greater than half that of the original stimulus compound St and
S g ; indeed, it may be even less. Our sixth corollary is accordingly
formulated as follows:

VI. If two stimulus aggregates, jointly and equally conditioned
to a reaction, are later presented separately , the reaction strength
evokable by each may be less than habit summation dynamics
would indicate, and in extreme cases may be even less than half
that of the compound originally conditioned, depending upon the
amount of afferent neural interaction effects in the original condi-
tioning situation. It will be noted that Corollary VI also consti-
tutes a limitation on the generality of Corollaries I and II, the
extent of the limitation depending on the amount of afferent inter-
action involved.

Presumptive critical illustrations of Corollary VT are found in
those cases of what Pavlov calls the “overshadowing” of one stim-
ulus component in a conditioned stimulus compound by another
component ( 6 , pp. 142-144). For example, in one case a stimulus
compound made up of a tone and three electric lamps was con-
ditioned to the salivary reaction so that the compound would evoke
eight drops of saliva in 30 seconds, though the lamps acting alone
evoked no secretion whatever. On the above assumptions the
lamps might very well have contributed materially to the joint
reaction of eight drops, yet have been so weakened by the influence
of afferent interaction under which conditioning took place as not
to pass the reaction threshold when presented separately. The
present set of hypotheses also demand that the tone when pre-
sented alone should have evoked less than eight drops in 30 sec-

COMPOUND CONDITIONED STIMULI 221

onds; unfortunately Pavlov does not give the results of this con-
trol test.

SUMMARY

Some writers, in an attempt to simplify the account of the learn-
ing process, have left the erroneous impression that the conditioned
stimulus, e.g., as represented by S in the symbol aH R , is a simple
or singular energy operating on one receptor end organ or, at most,
on a small number of such end organs of a single sense mode. In
actual fact an immense number of receptor end organs are involved
in every conditioning situation, however much it may have been
simplified by experimental methodology. Each stimulus “object”
represents a very complex aggregate of more or less alternative
potential stimulations, often extending into numerous receptor
modes.

Presumably every receptor which discharges an afferent impulse
during the conditioning process acquires an increment of habit
strength at each reinforcement. There is much reason to believe,
however, that the magnitude of the increments acquired by the
different receptor organs or receptor-organ aggregates varies
greatly. Among the factors which are believed to favor the acqui-
sition of large increments are: the dynamic or changing state of
the stimulus energy, the intensity of the stimulus energy, the
nature or “mode” of the receptor stimulated, the relative rarity of
the occurrence of the stimulus energy, and the chance that the
stimulus energy may previously have been conditioned to some
strongly emotional reaction.

Despite the great role that stimulus compounds play in adaptive
behavior, no unique primary principles have been found to be
operating. One major principle which seems to be active is that of
the summation of the habit-strength loadings borne by the several
stimulus elements or aggregates of a stimulus compound: The habit
strengths borne by the several stimulus components summate to
form a single effective habit strength which is equal to that which
would be produced by a number of consecutive standard reinforce-
ments equal to the sum of the number of reinforcements which
would be required to produce the separate habit strengths borne
by the several stimulus components. The action of the principle
of habit-strength summation is associated with a second primary
principle, the monotonic habit-reaction function. This is that the
strength of a reaction tendency, other things equal, is a monotonic

222

PRINCIPLES OF BEHAVIOR

increasing function of the effective strength of the habit, and so of
the reaction potential (p. 226 ff.) mediating it (Postulate 15, p.
344 ff.).

From these principles follows the first corollary, that the
strength of the reaction tendency evoked by two homogeneous con-
ditioned stimulus aggregates when acting in a stimulus compound
is less than the arithmetical sum of the two reaction tendencies
evokable by the components when acting separately. A second
corollary closely related to the first is that the joint reaction ten-
dency evoked by two stimulus components is greater than that
evokable by either component acting separately. Both of these
corollaries find experimental illustration under certain conditions.
A third corollary of the same group is to the effect that if a num-
ber of stimulus aggregates are equally conditioned to a reaction,
and varying numbers of them later simultaneously act on the recep-
tors of the organism, the greater the number of aggregate so acting,
other things equal, the greater will be the strength of the reaction
tendency thus evoked.

A third primary principle which, along with those just con-
sidered, is believed to be active in determining the dynamics of
stimulus compounds, is that known as afferent neural interaction.
This principle is to the effect that when afferent receptor discharges
occur at about the same time, they interact, changing each other
to varying degrees depending upon circumstances not as yet well
known (Postulate 2). Associated with the interaction principle is
the familiar principle of the primary stimulus generalization gradi-
ent (Postulate5).

From these latter primary principles there follow some addi-
tional corollaries. One of these is: In case a conditioned stimulus
aggregate is presented in a compound with a second or “extra”
stimulus aggregate, this stimulus compound, other things equal, will
evoke a weaker reaction tendency than will the stimulus aggregate
originally conditioned. An example of the action of this principle
is found in one type of Pavlovian external inhibition.

NOTES

The Equation for the Combination of the Habit Loadings Borne by the
Components of a Conditioned Stimulus Compound When
the Components Are Two in Number

This equation was derived by Dr. D. T. Perkins from an equivalent of the
combination of the habit-strength loadings borne by the elements of compound

COMPOUND CONDITIONED STIMULI

conditioned stimuli, in connection with the combination of two generalized habit
tendencies evokable by a particular stimulus compound (2). Recast in a form
appropriate for the present context, this equation is :

ij + = , x Hr + —

i x Hr X

(32)

where «i is one afferent element or aggregate of a conditioned stimulus compound
and s* is another such element or aggregate of the same conditioned stimulus
compound, H is the power of the stimulus, represented by si or s 2 , when com-
bined with suitable motivation to evoke the reaction (/?), and M is the physio-
logical limi t, of conditioning strength under the circumstances in which the condi-
tioning occurred.

It may easily be shown that the above equation is a special case of Day’s
general equation (30) for combining any number of habit strengths (XII, p. 200).

An E xam ple of the Combination of the Habit Strengths of Two Condi-
tioned Stimulus Elements or Aggregates by Means of Perkins’ Equation

In connection with Corollary V (p. 218) there was occasion to determine the
combined habit strengths of two conditioned stimulus elements or aggregates
delivered simultaneously as a stimulus compound, each element by hypothesis
carrying a habit loading of 40 habs. Because the exposition in the main text
is designed for non-mathematical readers, such combination values are usually
found by means of a table based on the simple growth function. The same out-
come could, however, have been secured in all cases by the use of Perkins' equa-
tion, as was actually done in the case of Corollary V. In that example ! x Hr —
40 habs and also 1,Hr = 40 habs. Accordingly, substituting these values in
Perkins’ equation (32), we have,

= 40 + 10- S2.X40

•» +

:,Br

80 -

80 -
64

1600

100

The Difficulty of Applying the Habit-Summation Equations in Quantita-
tive Detail to Concrete Behavior Situations

A minor difficulty in the way of applying the habit summation equations to
concrete behavior situations lies in the fact (p. 253 ff.) that habit must be com-
bined with a drive (D) or motivation before the stimulus can evoke a reaction.
If, however, both habit strength and reaction-evocation potential are on a centi-
grade scale, and the drive is chosen in such a way that its function when combined
multiplicatively with that of bHb is unity, no serious difficulty will arise. Since
it will not be possible to take up the matter of motivation until a later chapter,
nothing has been said about this, lest the reader be unnecessarily confused.

A major difficulty lies in the fact that a large number of the stimulus elements
always present in any conditioning situation can never be under direct and ready
•control of the investigator. As a result they can neither be administered sepa-

22 4

PRINCIPLES OF BEHAVIOR

rately from those deliberately employed by the investigator (and usually 01
primary interest to him) nor wholly eliminated from experimental situations in
which the experimenter seeks to determine the habit strength of certain stimulus
aggregates such as the light or the vibrator employed in the galvanic conditioning
situation (2) discussed above. If we let the aggregate habit loading of these
normally uncontrollable factors be represented by x, then instead of saying that
the galvanic reaction evoked by the light alone averaged 2.2 millimeters, we should
say that the action evoked by the light together with x unspecified and unmeasured
elements in the conditioned stimulus situation averaged 2.2 millimeters. In a similar
manner, we should say that the conditioned galvanic skin reaction evoked by the
joint action of the light, the vibrator, and x unspecified and unmeasured stimulus
elements averaged 5.1 millimeters. It accordingly comes about that in the

formula ^ ^ the aj-value appears twice in the denominator but only once in

the numerator, thus giving a smaller value to the ratio than it properly should have.

It might be supposed that the habit loading of these x stimulus elements
could be calculated by means of Day’s equation (XII, p. 200). As a matter of
fact this probably would be possible if we knew the value of the constant, M.
On the other hand, the value of M could be calculated if we only knew the habit
loading of the x stimulus elements. Perhaps the most promising possibility of
escaping from this dilemma lies in the determination of the value of the constant
M by some quite independent procedure. Meanwhile a considerable n um ber of
quasi-quantitative but empirically testable theorems may be derived from the
equations even under present conditions. Three of these have been outlined

above as corollaries based mainly on the hypothesis upon which both equations
depend.

Finally, the conjecture may be appended that possibly these x stimulus
elements, such as the cutaneous stimulations normally resulting from the contact
of the organism with its support and the multitude of proprioceptive stimulus
elements resulting from the posture of the organism, possess a relatively weak
capacity for acquiring conditioned habit loadings owing to the fact that since
they are more or less ubiquitous they must become conditioned in every condi-
tioning situation the organism encounters. Unless these situations become very
highly patterned on the stimulus side, it follows that stimulus elements which are

conditioned to all kinds of reactions would ultimately become permanently
extinguished and thus finally i mm une to any further conditioning. This of course
would hold for usual or customary postures of the organism, but not for rare or
unusual postures. There is great need for experimental research in this field,
but the problem is a difficult one.

REFERENCES

1. Hull, C. L. The problem of stimulus equivalence in behavior theory.

Psychol. Rev., 1939, 46, 9-30.

2. Hull, C. L. Explorations in the patterning of stimuli conditioned to the

GS.R. J. Exper. Psychol., 1940, 27, 95-110.

3. Hull, C. L. Conditioning: Outline of a systematic theory of learning.

Chapter II in The psychology of learning (forty-first yearbook. Na-
tional Society for the Study of Education). Bloomington, 111.: Public
School Publishing Co., 1942.

COMPOUND CONDITIONED STIMULI 225

4. Kohler, W. Gestalt psychology. New York: Liveright, 1929.

5. Lashley,^ K. S. The mechanism of vision: XV. Preliminary studies of

the rat’s capacity for detail vision. J. Gen. Psychol., 1938, 18, 123-193.

6 . Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

7. Shepard, J. F., and Fogelsanger, H. M. Studies in association and inhibi-

tion. Psychol. Rev., 1913, 20, 290-311.

CHAPTER XIV

Motivation and Reaction Potential

It may be recalled that when the problem of primary rein-
forcement was under consideration (p. 68 ff.), the matter of or-
ganic need played a critical part in that the reduction of the need
constituted the essential element in the process whereby the reac-
tion was conditioned to new stimuli. We must now note that the
state of an organism’s needs also plays an important role in the
causal determination of which of the many habits possessed by an
organism shall function at a given moment. It is a matter of com-
mon observation that, as a rule, when an organism is in need of
food only those acts appropriate to the securing of food will be
evoked, whereas when it is in need of water, only those acts appro-
priate to the securing of water will be evoked, when a sexual hor-
mone is dominant only those acts appropriate to reproductive
activity will be evoked, and so on. Moreover, the extent or inten-
sity of the need determines in large measure the vigor and persist-
ence of the activity in question.

By common usage the initiation of learned, or habitual, pat-
terns of movement or behavior is called motivation. The evocation
of action in relation to secondary reinforcing stimuli or incentives
will be called secondary motivation; a brief discussion of incentives
was given above (p. 131) in connection with the general subject of
amount of reinforcement. The evocation of action in relation to
primary needs will be called primary motivation; this is the subject
of the present chapter.

THE EMPIRICAL* ROLES OF HABIT STRENGTH AND DRIVE IN THE

DETERMINATION OF ACTION

Casual observations such as those cited above often give us
valuable clues concerning behavior problems, but for precise solu-
tions, controlled quantitative experiments usually are necessary.
In the present context we are fortunate in having an excellent em-
pirical study which shows the functional dependence of the per-
sistence of food-seeking behavior jointly on (1) the number of
reinforcements of the habit in question, and (2) the number of

226

PRIMARY MOTIVATION AND REACTION POTENTIAL 227

hours of food privation. Perm ( 12 ) and Williams ( 20 ) trained
albino rats on a simple bar-pressing habit of the Skinner type
(p. 87), giving separate groups different numbers of reinforce-
ments varying from 5 to 90 under a standard 23 hours’ hunger.
Later the groups were subdivided and subjected to experimental

Fio. 48. Column diagram of the Perin-Williams data showing quantita-
tively how the resistance to experimental extinction in albino rats varies
jointly with the number of reinforcements and the number of hours of food
privation at the time the extinction occurred. The cross-hatched columns
rejjresent the groups of animals reported by Williams (SO ) ; the non-hatched
columns represent the groups reported by Perin. (Figure reproduced from
Perin, IS, p. 106.)

extinction 1 with the amount of food privation varying from 3 to
22 hours.

The gross outcome of this experiment is shown in Figure 48,
where the height of each column represents the relative mean num-
ber of unreinforced reactions performed by each group before ex-
perimental extinction yielded a five-minute pause between succes-
sive bar pressures. The positions of the twelve columns on the
base shows clearly the number of reinforcements and the number

1 For an account of experimental extinction, see pp. 258 ff.

Fio. 49. Graphic representation of the data showing the systematic rela-
tionship between the resistance to experimental extinction (circles) and the
number of hours’ food privation where the number of reinforcements is
constant at 16. The smooth curve drawn through the sequence of circles
represents the slightly positively accelerated function fitted to them. This
function is believed to hold only up to the number of hours of hunger em-
ployed in the original habit formation process: in the present case, 23. (Fig-
ure adapted from Perin, 12 , p. 104.)

Fid. 50. Graphic representation of the two “learning” curves of Figure 48,
shown in the same plane to facilitate comparison. The solid circles represent
the empirical values corresponding to the heights of the relevant columns
of Figure 48; the one hollow circle represents a slightly interpolated value.
The smooth curves drawn among each set of circles represent the simple
growth functions fitted to each set of empirical data. (Figure adapted from
Perm, 12 , p. 101.)

228

PRIMARY MOTIVATION AND REACTION POTENTIAL 229

of hours’ food privation which produced each. It is evident from
an examination of this figure that both the number of reinforce-
ments and the number of hours of food privation are potent factors
in determining resistance to experimental extinction. Moreover,
it is clear that for any given amount of food privation, e.g., 3 or 22
hours, the different numbers of reinforcements yield a close approxi-
mation to a typical positive growth function. On the other hand,

Fio. 51. Three-dimensional graph representing the fitted “surface” corre-
sponding quantitatively to the action of the number of reinforcements and the
number of hours of food privation following satiation, in the joint determina-
tion of the number of unreinforced acts of the type originally conditioned
which are required to produce a given degree of experimental extinction.
(Figure adapted from Perm, 12 , p. 108.)

it is equally clear that for a given number of reinforcements, e.g.,
16, the number of hours of food privation has an almost linear
functional relationship to the resistance to experimental extinction.

For a more precise analysis of thes? functional relationships it
is necessary to fit two-dimensional curves to the data. The results
of this procedure are presented in Figures 49 and 50. Figure 49
shows that resistance to extinction at the 16-reinforcement level is
a slightly positively accelerated function of the number of hours’
food privation for the first 22 hours. Figure 50 shows that a posi-

2 3 o

PRINCIPLES OF BEHAVIOR

tive growth function fits both “learning” curves fairly well. An
examination of the equations w r hich generated these curves reveals
that the asymptotes differ radically, clearly being increasing func-

D A V S

Fio. 52. Graph showing the relation-
ship of the action potentiality as a
function of the length of food priva-
tion following satiation. First note the
fact that there is an appreciable
amount of action potentiality at the
beginning of this graph, where the
amount of food privation is zero. Next,
observe that the curve is relatively
high at one day of food privation,
which was the degree of drive under
which the original training occurred.
Finally, note that the rise in action
potentiality is fairly continuous up to
about five days, after which it falls
rather sharply. This fall is evidently
due to exhaustion, as the animals died
soon after. The function plotted as
the smooth curve of Figure 49 corre-
sponds only to the first section of the
present graph and clearly does not rep-
resent the functional relationship be-
yond a point where the number of
hours of food privation is greater than
23. (Figure reproduced from Skinner,
IS, p. 396.)

tions of the number of hours of
food privation, but that the
rates at which the curves ap-
proach their respective asymp-
totes are practically identical
(F equals approximately 1/25
in both cases). Finally it may
be noted that both curves,
when extrapolated backward to
where the number of reinforce-
ments would equal zero, yield a
negative number of extinctive
reactions amounting to approx-
imately four. This presumably
is a phenomenon of the reac-
tion threshold which will be
discussed in some detail later
(p. 322) ; it is believed to mean
that a habit strength sufficient
to resist four extinction reac-
tions is necessary before reac-
tion will be evoked by the
stimuli involved.

For a final examination of
the outcome of the experiment
as a whole, the curves shown in
Figures 49 and 50 were syn-
thesized in such a way as to
yield a surface fitted to the
tops of all the columns of Fig-
ure 48. This surface is shown
in Figure 51. An examination
of this figure reveals the impor-
tant additional fact that when

the surface is extrapolated to where the number of hours’ food pri-
vation is zero, the resistance to experimental extinction presumably
will still show a positive growth function with n-values of consid-
erable magnitude. As a matter of fact, the asymptote of the growth

PRIMARY MOTIVATION AND REACTION POTENTIAL 231

function where h = 0 (satiation) is 28 per cent of that where h =
22 hours.

These last results are in fairly good agreement with comparable
values from several other experimental studies. Measurements of
one of Skinner’s published graphs, reproduced as Figure 52, indicate
that his animals displayed approximately 17 per cent as much food-
seeking activity at satiation as at 25 hours’ food privation. Finch
(3) has shown that at satiation a conditioned salivary reaction in
nine dogs yielded a mean of 24 per cent as much secretion as was
yielded at 24 hours’ food privation. Similarly, Zener (22) reports
that the mean salivary secretion from four dogs average at satia-
tion 24 per cent as much as at from 21 to 24 hours’ food privation.
The considerable amounts of responsiveness to the impact of con-
ditioned stimuli when the organism is in a state of food satiation
may accordingly be considered as well established.

The continued sexual activity of male rats for some months
after castration points in the same direction. Stone (15) reports
that male rats which have copulated either shortly before or shortly
after castration, when an adequate supply of hormone would be
present, continue to show sexual behavior sometimes as long as
seven or eight months after removal of the testes. According to
Moore et al. (10), Stone (15), and Beach (1), this operation re-
moves within 20 days not only the source of testosterone but,
through the resulting atrophy of accessory glands, also the source
of other specifically supporting secretions. A few weeks after cas-
tration, therefore, when the normal supply of sex hormones in the
animal’s body has been exhausted, the sex drive is presumably in
about the same state as is the food drive after complete food
satiation. The continued sexual activity of these animals thus pre-
sents a striking analogy to the continued operation of the food-
release bar by Perm’s rats after food satiation. While not abso-
lutely convincing, this evidence from the field of sexual behavior
suggests that the performance of learned reactions to moderate
degrees in the absence of the specific drive involved in their origi-
nal acquisition may be sufficiently general to apply to all primary
motivational situations.

Closely related to this same aspect of Perm’s investigation is
a study reported by Elliott ( 2 ). Albino rats were trained in a maze
under a thirst drive with water as the reinforcing agent until the
true path was nearly learned, when the drive was suddenly shifted
to hunger and the reinforcing agent to food. The outcome of this

232 PRINCIPLES OF BEHAVIOR

procedure is shown in Figure 53. There it may be seen that on the
first trial under the changed condition of drive there was an appre-
ciable disturbance of the behavior in the form of an increase in
locomotor time; there was also an increase, of about the same
proportion, in blind-alley entrances. On the later trials, however,

Fiq. 53. Graphs showing the disruptive influence on a maze habit set up
in albino rats on the basis of a water reinforcement, of having the drive (on
the tenth day) suddenly shifted from thirst to hunger. (Reproduced from
Elliott, g, p. 187.)

the learning process appeared to proceed much as if no change
had been made in the experimental conditions.

As a final item in this series there may be mentioned an em-
pirical observation of Pavlov concerning the effect on an extin-
guished conditioned reaction of increasing the drive. On the anal-
ogy of Perm’s experiment, it might be expected that this would
again render the reaction evocable by the stimulus; and this m
fact took place. In this connection Pavlov remarks {11, p. 127):

PRIMARY MOTIVATION AND REACTION POTENTIAL 233

To illustrate this last condition
we may take instances of differen-
tial inhibitions established on the
basis of an alimentary reflex. If,
for example, the dog has been
kept entirely without food for a
much longer period than usual be-
fore the experiment is conducted,
the increase in excitability of the
whole alimentary nervous mech-
anism renders the previously es-
tablished differential inhibition
wholly inadequate.

EMPIRICAL DIFFERENTIAL
REACTIONS TO IDENTICAL
EXTERNAL ENVIRONMENTAL
SITUATIONS ON THE BASIS
OF DISTINCT DRIVES

A second important type
of motivational problem was
broached in an experiment re-
ported by Hull (d) . Albino rats
were trained in the rectangular
maze shown in Figure 54. On
some days a given animal
would be run in the maze
when satiated with water, but
with 23 hours’ food privation,
whereas on other days the same
rat would be run when satiated
with food but with 23 hours of
water privation. The two types
of days alternated according to
a predetermined irregularity.
On the food-privation days the
reinforcement chamber always
contained food and the left en-

Fio. 54. Diagram of the maze em-
ployed in Hull’s differential drive ex-
periment. S= starting chamber; G
— food chamber; D', D " = doors man-
ipulated by cords from the experiment-
er’s stand; B', B ” = barriers across pas-
sageway, one of which was always
closed. The course pursued by a typ-
ical rat on a “false” run is shown by
the sinuous dotted line. Note that the
animal went down the “wrong” side of
the maze far enough to see the closed
door at B* and then turned around.
(Reproduced from Hull, 6.)

trance, say, to the chamber was blocked so that access could be had
only by traversing the right-hand side of the rectangle. On the
water-privation days the reinforcement chamber always contained
water, and the right-hand entrance to the reinforcement chamber
would be blocked so that access to the water could be had only by

PRINCIPLES OF BEHAVIOR

2 34

traversing the left-hand side of the rectangle. The outcome of this
experiment is shown in Figure 55. There it may be seen that while
learning was very slow, the animals of the experimental group
gradually attained a considerable power of making the reaction
which corresponded to the drive dominant at the time.

The capacity of rats to learn this type of discrimination was
later demonstrated more strikingly by Leeper ( 8 ), in a substantially
similar investigation. Leeper’s experiment differed, however, in the

hunger and thirst motivation (5, p. 263).

detail that two distinct reinforcement chambers were employed and
no passageways were blocked at any time, so that if on a “food”
day the rat went to the water side he always found water, and if
on a “water” day he went to the food side, he always found food.
Under these conditions the animals learned to perform the motiva-
tional discrimination with great facility; Leeper’s animals needed
only about one-twelfth the number of trials required by the origi-
nal Hull technique, though again the process of acquisition was

gradual. 1

iThis striking difference is attributed in part to the operation of spatial
orientation and in part to the fact that when rats are deprived of either
food or water they do not consume a normal amount of the other sud-

PRIMARY MOTIVATION AND REACTION POTENTIAL 235

DOES THE PRINCIPLE OF PRIMARY STIMULUS-INTENSITY

GENERALIZATION APPLY TO THE DRIVE STIMULUS ( S D )?

A factor with considerable possible significance for the under-
standing of motivation is the relationship between the degree of
similarity of the need at the time of reinforcement and that at the
time of extinction, on the one hand, and the associated resistance
to experimental extinction on the other. No specific experiments
have been found bearing exactly on this point, but several inciden-
tal and individually inconclusive bits of evidence may be men-
tioned as indicating the general probabilities of the situation.

The first of these was reported by Heathers and Arakelian U).
Albino rats were trained to secure food pellets by pressing a bar
in a Skinner-Ellson apparatus. Next, half of the animals were par-
tially extinguished under a weak hunger, and the remainder were
extinguished to an equal extent under a strong hunger. Two days
later the animals were subjected to a second extinction, half of
each group under the same degree of hunger as in the first extinc-
tion, and the remaining half under a drive equal to the first-extinc-
tion hunger of the other group. Combining the state of food priva-
tion of the first and second extinctions, there were thus four hunger-
extinction groups:

1, strong-strong; 2, strong-weak; 3, weak-strong; 4, weak-weak.

The authors report that a statistical pooling of the results from
these four groups of animals revealed a tendency of the rats extin-
guished twice on the same drive to resist extinction less than did
those animals which were extinguished the second time on a drive
different from that employed on the first occasion. In two inde-
pendent studies this difference amounted to approximately 4 and
6 per cent respectively; the latter results are reported to have a
probability of 8 in 10 that the difference was not due to chance.
This experimental outcome is evidently related to the primary
generalization of stimulus intensity and suggests that perseverative

stance; this prevents genuine satiation of the supposedly satiated drive.
For example, thirsty rats supposedly satiated with food will, after receiving
even a few drops of water, very generally eat if food is available ( 6 , p. 270) ;
and rats, like humans, frequently drink while eating dry food if water is
available. Thus after the first trial Leeper’s animals were presumably oper-
ating under both drives, and one drive or the other was reinforced no mat-
ter which path was traversed.

PRINCIPLES OF BEHAVIOR

236

extinction effects are to some extent specific to the primary drive
or need intensity under which the extinction occurs.

By analogy, the stimulus-intensity generalization gradient ap-
parently found in the case of extinction effects just considered
strongly suggests the operation of the same principle in the case
of reinforcement effects. Now, such a gradient has been demon-
strated experimentally by Hovland (see p. 186 ff.) ; it naturally
has its greatest value (Figure 43) at the point of reinforcement.
Consequently it is to be expected that in a curve of motivation
intensity such as that of Skinner (shown in Figure 52) a special
elevation or inflection would appear at the drive intensity at which
the original reinforcement occurred. Whether a mere coincidence
arising from sampling errors or not, exactly such an inflection may
be seen in Skinner’s empirical graph at one day of food privation,
which was in fact the drive employed by Skinner in the training of
the animals in question. The present set of assumptions implies
that if Skinner’s curve as shown in Figure 52 were to be plotted
in detail by hours rather than days, it would present a positive
acceleration from zero to one day of food privation. Now, Perin’s
study did plot this region in some detail, and Figure 49 shows that
a positively accelerated function was found. These facts still fur-
ther increase the probability that the principle of stimulus-intensity
generalization applies to the drive stimulus ( S D ).

THE INFLUENCE OF CERTAIN DRUGS ON EXPERIMENTAL
EXTINCTION AND ITS PERSEVE RATIONAL EFFECTS

Certain drugs are known to influence markedly the phenomena
of experimental extinction. Switzer (18) investigated the effect of
caffeine citrate on the conditioned galvanic skin reaction in human
subjects, using a control dose of milk sugar. He found that caf-
feine increased resistance to experimental extinction; incidentally
he also found that caffeine increased the amplitude of the uncon-
ditioned galvanic skin reaction and decreased the reaction latency.

Pavlov (11, p. 127) reported a somewhat related experiment
performed by Nikiforovsky. An alimentary salivary conditioned
reflex had been set up to a tactile stimulus on a dog’s forepaw.
This reaction tendency generalized to other parts of the animal’s
skin including a point on the back which subsequently was com-
pletely extinguished. At the latter stage of training the stimulus
on the paw yielded five drops of saliva during the first minute of

PRIMARY MOTIVATION AND REACTION POTENTIAL 237

stimulation, whereas stimulation of the extinguished spot on the
back yielded a zero reaction. Thereupon, the animal was given a
subcutaneous injection of 10 c.c. of 1 per cent solution of caffeine.
A few minutes later the stimulus when applied to the forepaw
evoked four drops during the first minute, and when applied to
the previously extinguished spot on the back, yielded three drops
( 11 , p. 128), thus indicating a major dissipation of the extinction
effects.

Miller and Miles ( 9 ) have contributed to this field. They
demonstrated in albino rats traversing a 25-foot straight, enclosed
runway that an injection of caffeine sodio-benzoate reduced the
locomotor retardation due to experimental extinction by about
two-thirds. In the same study it was 6hown that the retardation
in locomotor time due to satiation was reduced by the caffeine
solution approximately one-half (£).

Benzedrine is another substance which when thrown into the
blood stream has the power of greatly retarding the onset of experi-
mental extinction. This was demonstrated by Skinner and Heron
( 14 ) to hold for the Skinner bar-pressing habit.

SEX HORMONES AND REPRODUCTIVE ACTIVITY

As a final set of empirical observations concerning motivation
we must consider briefly the relation of sex hormones to repro-
ductive behavior. Within recent years an immense amount of
excellent experimental work has been performed in this field, though
only brief notice of it can be taken in this place. An account of
two typical bits of this work was given above (Figures 11 and 12).
In a recent comprehensive summary by Beach ( 1 ) the following
propositions appear to have fairly secure empirical foundation:

1. Animals of practically all species which through castration have
become sexually unresponsive to ordinary incentive stimulation, become
responsive promptly on the injection of the appropriate hormone — usu-
ally testosterone proprionate for males and estrogen for females.

2. Presumptively normal male rats differ greatly in their sexual re-
sponsiveness, all the way from those which will attempt copulation with
inanimate objects to those which will not react even to an extremely re-
ceptive and alluring female. The injection of testosterone usually raises
the reactivity of all but a few of the most sluggish animals. Alterna-
tively, the presentation of an especially attractive incentive tends to
have the same objective effect, though to a lesser degree ( 17 ).

3. Destruction of the cerebral cortex decreases sexual reactivity roughly
in proportion to the extent of such destruction, very much as occurs in

PRINCIPLES OF BEHAVIOR

238

the case of food habits. If destruction has not been too great, injection
of the hormone will largely restore sexual responsiveness to appropriate
incentives. The presentation of an exceptionally attractive incentive will,
however, have much the same effect upon the objective behavior of such
organisms.

4. Virgin male organisms which are unresponsive to an ordinary re-
ceptive female, after a few copulations under the influence of an injec-
tion of the hormone will remain responsive long after the hormone has
presumably disappeared from the animal’s body. This is believed to be
caused by the learning resulting from the incidental reinforcement which
occurred when the animal was under the influence of the hormone (I).

5. Many intact individuals of both sexes in most species occasionally
manifest a portion of the behavior pattern characteristic of the oppo-
site sex. Injection of the sex hormone of the opposite sex in castrated
individuals of either sex tends strongly to the evocation of the sexual be-
havior pattern characteristic of normal organisms of the opposite sex on
appropriate stimulation; this, however, is not usually as complete as the
gross anatomical equipment of the organisms would seem to permit. Curi-
ously enough, large doses of testosterone given to male rats make possible
the elicitation of all elements of the typical female sexual behavior (I).

PRIMARY MOTIVATIONAL, CONCEPTS

With the major critical phenomena of primary motivation 1
now before us, we may proceed to the attempt to formulate a theory
which will conform to these facts.

At the outset it will be necessary to introduce two notions not
previously discussed. These new concepts are analogous to that
of habit strength (sHr) which, it will be recalled (p. 114), is a
logical construct conceived in the quantitative framework of a cen-
tigrade system.

The first of the two concepts is strength of primary drive; this
is represented by the symbol D. The strength-of-drive scale is
conceived to extend from a zero amount of primary motivation
(complete satiation) to the maximum possible to a standard organ-
ism of a given species. In accordance with the centigrade principle
this range of primary drive is divided into 100 equal parts or units.
For convenience and ease of recall, this unit will be called the mote,
a contraction of the word motivation with an added e to preserve
normal pronunciation.

Because of the practical exigencies of exposition the second of

J The empirical phenomena of secondary motivation, including such mat-
ters as incentive (p. 131 ff.), fractional anticipatory goal- and subgoal-reao-
tions, cannot be treated in the present volume because space is not available.

PRIMARY MOTIVATION AND REACTION POTENTIAL 239

the new concepts has already been utilized occasionally in the last
few pages, where it has been referred to as the “reaction tendency,”
a term in fairly general use though lacking in precision of meaning.
For this informal expression we now substitute the more precise
equivalent, reaction- evocation 'potentiality ; or, more briefly, reac-
tion potential. This will be represented by the symbol g E R .
Like habit ( 3 H R ) and drive (D), reaction-evocation potential is
also designed to be measured on a 100-point scale extending from
a zero reaction tendency up to the physiological limit possible to
a standard organism. The unit of reaction potentiality will be
called the wat, a contraction of the name Watson.

It should be evident from the preceding paragraphs that D and
b E b are symbolic constructs in exactly the same sense as B H R (see
p. Ill ff.), and that they share both the advantages and disadvan-
tages of this status. The drive concept, for example, is proposed
as a common denominator of all primary motivations, whether due
to food privation, water privation, thermal deviations from the
optimum, tissue injury, the action of sex hormones, or other causes.
This means, of course, that drive will be a different function of
the objective conditions associated with each primary motivation.
For example, in the case of hunger the strength of the primary drive
will probably be mainly a function of the number of hours of food
privation, say; in the case of sex it will probably be mainly a func-
tion of the concentration of a particular sex hormone in the ani-
mal’s blood; and so on. Stated formally,

D = f(h)

D = /(c)

D = etc.,

where h represents the number of hours of food privation of the
organism since satiation, and c represents the concentration of a
particular hormone in the blood of the organism.

Turning now to the concept of reaction-evocation potentiality,
we find, thanks to Perm’s investigation sketched above (p. 227 ff.),
that we are able at once to define 3 E R as the product of a function
of habit strength ( 3 H R ) multiplied by a function of the relevant
drive ( D ). This multiplicative relationship is one of the greatest
importance, because it is upon g E R that the amount of action in
its various forms presumably depends. It is clear, for example,
that it is quite impossible to predict the vigor or persistence of a
given type of action from a knowledge of either habit strength or

240

PRINCIPLES OF BEHAVIOR

drive strength alone; this can be predicted only from a knowledge
of the product of the particular functions of 8 Hr and D respec-
tively; in fact, this product constitutes the value which we are
representing by the symbol b Er,

SUMMARY AND PRELIMINARY PHYSIOLOGICAL INTERPRETATION

OF EMPIRICAL FINDINGS

Having the more important concepts of the systematic approach
of primary motivation before us, we proceed to the formulation of
some empirical findings as related to motivation.

Most, if not all, primary needs appear to generate and throw
into the blood stream more or less characteristic chemical sub-
stances, or else to withdraw a characteristic substance. These
substances (or their absence) have a selective physiological effect
on more or less restricted and characteristic portions of the body
(e.g., the so-called “hunger” contractions of the digestive tract)
which serves to activate resident receptors. This receptor activa-
tion constitutes the drive stimulus, S D (p. 72 ff.). In the case of
tissue injury this sequence seems to be reversed; here the energy
producing the injury is the drive stimulus, and its action causes
the release into the blood of adrenal secretion which appears to be
the physiological motivating substance.

It seems likely, on the basis of various analogies, that, other
things equal, the intensity of the drive stimulus would be some form
of negatively accelerated increasing function of the concentration
of the drive substance in the blood. However, for the sake of ex-
pository simplicity we shall assume in the present preliminary
analysis that it is an increasing linear function.

The afferent discharges arising from the drive stimulus (Sz>)
become conditioned to reactions just the same as any other elements
in stimulus compounds, except that they may be somewhat more
potent in acquiring habit loadings than most stimulus elements or
aggregates. Thus the drive stimulus may play a role in a con-
ditioned stimulus compound substantially the same as that of any
other stimulus element or aggregate (p. 74 ff.). As a stimulus, Sd
naturally manifests both qualitative and intensity primary stimulus
generalization in common with other stimulus elements or aggre-
gates in conditioned stimulus compounds (p. 185 ff.).

It appears probable that when blood which contains certain
chemical substances thrown into it as the result of states of need,

PRIMARY MOTIVATION AND REACTION POTENTIAL 241

or which lacks certain substances as the result of other states of
need, bathes the neural structures which constitute the anatomical
bases of habit ( 5 //#), the conductivity of these structures is aug-
mented through lowered resistance either in the central neural tissue
or at the effector end of the connection, or both. The latter type of
action is equivalent, of course, to a lowering of the reaction
threshold and would presumably facilitate reaction to neural im-
pulses reaching the effector from any source whatever. As Beach
U) suggests, it is likely that the selective action of drives on par-
ticular effector organs in non-leamed forms of behavior acts mainly
in this manner. It must be noted at once, however, that sensitizing
a habit structure does not mean that this alone is sufficient to
evoke the reaction, any more than that caffeine or benzedrine alone
will evoke reaction. Sensitization merely gives the relevant neural
tissue, upon the occurrence of an adequate set of receptor dis-
charges, an augmented facility in routing these impulses to the
reactions previously conditioned to them or connected by native
(inherited) growth processes. This implies to a certain extent the
undifferentiated nature of drive in general, contained in Freud’s
concept of the “libido.” However, it definitely does not presup-
pose the special dominance of any one drive, such as sex, over the
other drives.

While all drives seem to be alike in their powers of sensitizing
acquired receptor-effector connections, their capacity to call forth
within the body of the organism characteristic and presumably dis-
tinctive drive stimuli gives each a considerable measure of distinc-
tiveness and specificity in the determination of action which, in
case of necessity, may be sharpened by the process of patterning
(see p. 349 ff.) to almost any extent that the reaction situation
requires for adequate and consistent reinforcement. In this respect,
the action of drive substances differs sharply from that of a pseudo-
drive substance such as caffeine, which appears to produce nothing
corresponding to a drive stimulus.

Little is known concerning the exact quantitative functional
relationship of drive intensity to the conditions or circumstances
which produce it, such as the number of hours of hunger or the
concentration of endocrine secretions in the blood. Judging from
the work of Warden and his associates (19 ) , the relationship of
the hunger drive up to two or three days of food privation would
be a negatively accelerated increasing function of time, though a
study by Skinner (Figure 52) suggests that it may be nearly linear

PRINCIPLES OF BEHAVIOR

242

up to about five days. For the sake of simplicity in the present
explorational analysis we shall assume the latter as a first approxi-
mation.

Physiological conditions of need, through their sensitizing action
on the neural mediating structures lying between the receptors and
the effectors ( 8 H R ), appear to combine with the latter to evoke
reactions according to a multiplicative principle, i.e., reaction-
evocation potentiality is the product of a function of habit strength
multiplied by a function of the strength of drive:

sEr — KsH r ) X

In the next section it shall be our task to consider in some detail
what these functions may be; if successful we shall then possess
the main portion of a molar theory of primary motivation.

THE QUANTITATIVE DERIVATION OF bEr FROM bBr AND D

Since we have taken Perm's experiment as our main guide in
the analysis of the primary motivational problem in general, it
will be convenient to take the need for food as the basis for the
detailed illustration of the working of the molar theory of motiva-
tion; this we now proceed to develop.

Turning first to the habit component of g E R , we calculate the
values of b Hr as a positive growth function; we use in this cal-
culation the fractional incremental value ( F ) found by Perin to
hold for the learning processes represented in Figure 50, which
was approximately 1 /25 for each successive reinforcement. On this
assumption the values at various numbers of reinforcements, e.g.,
0, 1, 3, 9, 18, 36, and 72, have been computed. These are shown
in column 2 of Table 5.

The habit-strength values of column 2, Table 5, consist of the
physiological summation of the habit-strength loadings of the stim-
ulus components, represented by the original drive stimulus Sd
and the non-drive components, which we shall represent by St.
Assuming as a matter of convenience that So' and Sj have equal
loadings, the value of each (see fifth terminal note) is easily cal-
culated for the several numbers of reinforcements. These values

are shown in column 3 of Table 5.

Turning next to the matter of drive, it will be assumed that
the original learning took place under a 24-hour food privation.
Assuming further that drive is a linear function of the number o

TABLE 5

Table Showing the Preliminary Steps in the Derivation op a Series op Theoretical Reaction-Potential Values from
Varied Set op Antecedent Reinforcements Under a Drive of 20 Units’ Strength, the Resulting Habits Being Evaluated

< £*

PRINCIPLES OF BEHAVIOR

244

hours’ hunger and that (Figure 52) the maximum of 100 motes
would be reached at five days or 120 hours, Perin’s periods of food
privation may be converted into units of drive strength by multi-
plying the number of hours’ food privation by the fraction 100/120.
In this way we secure the following drive or D-values:

Number of hours’ food privation ( h ) : 0 3 8 16 24

Strength of drive in motes (D) : 0 2.5 6.667 13.333 20

Deviation (d) of possible D’s from the

drive (D y ) of original learning: 20 17.5 13.333 6.667 0

Now, S D is assumed to be approximately a linear function of D.
It follows from this and the principle of primary stimulus general-
ization that action evoked under any other intensity of drive (and
drive stimulus) than that involved in the original habit formation
must be subject to primary intensity-stimulus generalization. As-
suming the relatively flat gradient yielded by an F-value of 1/50,

it is easy to calculate the value of b d Hr (p. 199 ff.) at each degree

of the five D-values taken above. These bdHr values are shown
in columns 4, 5, 6, 7, and 8 respectively of Table 5. A glance at
the bottom entries of each of the columns shows that the values
of 8 d H r fall progressively from 46.34 at D = 20 (i.e., d = 0.00)
to 30.94 at D = 0 (i.e., d = 20) .

We must now combine these habit values by the process of
physiological summation characteristic of conditioned stimulus
compounds (p. 223 ff.) (neglecting the effects of afferent interac-
tion) with the habit loading of the non-drive stimulus component
of the compound which is represented by the values appearing m
column 3. The physiological summation of the values in column 3
with the values of columns 4 to 8 gives us the habit-strength values
shown in columns 9, 10, 11, 12, and 13 of Table 5. It will be
noticed that this final recombination of the sHr values where
D = 20 yields exactly the same values as those of column 2. This
is because when reaction evocation occurs at the original drive
( D'), i.e., where D = D ', no distortion of the S D component of the
habit results, the synthesis being exactly the reverse of the analysis

which took place between columns 2 and 3.

With the theoretical values of j( s H R ) available in columns 9
to 13 inclusive of Table 5, we may now turn our attention to the
problem of /(D). It is assumed that D itself acts upon a H B as a
direct proportion. However, there is the complication that other

PRIMARY MOTIVATION AND REACTION POTENTIAL 245

or alien drives active at the time (represented in the aggregate by
the symbol D) have the capacity to sensitize habits not set up in
conjunction with them. Let it be supposed that this generalized
effect of alien drives adds 10 points to the actual drive throughout

the present situation. Thus the effective drive (D) operative on a
given habit would necessarily involve the summation of D and D*

m t 1 h f. cas ® of the 24 ' hour food privation a simple summation
would in the present situation amount to 10 + 20 or 30 and at

120 hours it would be 10 + 100, or 110. In order to maintain our

centigrade system the simple summation must be divided by the

maximum possible under these assumptions, or 110. Accordingly
we arrive at the formula,

D = 100 - P + D .

_ D + 100

where D represents the effective drive actually operative in pro-
ducing the reaction potential.

Now, assuming that reaction evocation potentiality is essen-
tially a multiplicative function of habit strength and drive, i.e
that, 1 ' *

sE r = J( s Hb) X /(D),

since / (sffjj) is a H R , and /(D) is D, we have by substitution,

sE R = sH r X D.

However, since both B H R and D are on a centigrade scale, their

simple product would yield values on a ten-thousand point scale;

therefore, to keep a E R also to a centigrade scale we write the equa-
tion,

sHr X D
100

Substituting the equivalent of D and simplifying, we have as our
final equation,

sEr

Tr D 4- D

stlR-z

D + 100

The second portion of this formula, with the various D values sub-
stituted, is,

10 + 0 . 10 4- 2.5.

HO 110

10 4- 6.667
110

10 4- 13.333 .
110

10 4- 20
110

- .0909

.1136

.1515

.2121

.2727.

PRINCIPLES OF BEHAVIOR

24 6

The values of b Er are accordingly obtained simply by multiply-
ing the several entries of column 9 by .0909, those of column 10
by .1136, and so on. These products are presented in detail in
columns 14, 15, 16, 17 and 18 of Table 5, which are the values we
have been seeking; they are shown diagrammatically by the curved
surface of Figure 56. A comparison of the theoretical values of
Figure 56 with the surface fitted to the empirical values represented

NUMBER OF REINFORCEMENTS (N)

Fia. 56. Graphic representation of the theoretical joint determination of
reaction potential by various numbers of reinforcements under a drive ( D ’)
of 20 units’ strength when functioning under drives (Z)) of various strengths
less than that of the original habit formation. Note the detailed agreement
with the comparable empirical results shown in Figure 51.

by the circles in Figure 51 indicates that the theoretical derivations
approximate the facts very closely indeed.

Computations analogous to the preceding have shown that the
present set of postulates and constants also hold when D > D' at
least up to three days of food privation. The theoretical curve
for all values of D between 0 and 72 hours yields a positively
accelerated reaction potential up to 24 hours ( D ' in the present
analysis), where there is a slight inflection; as D increases above
D' there is at first a brief period of positive acceleration, which is

PRIMARY MOTIVATION AND REACTION POTENTIAL 247

followed by a protracted period that is nearly linear, the whole
showing a fair approximation to Figure 52.

Generalizing from Table 5 and Figure 56, the following corol-
laries may be formulated as a kind of condensed summary of
the implications of the present set of assumptions as shown by the
preceding computations:

I. When habit strength is zero , reaction-evocation potential is
zero.

II. When primary drive strength ( D ) is zero, reaction- evocation
potential ( 8 E R ) has an appreciable but relatively low positive value
which is a positive growth junction of the number of reinforce-
ments. Corollaries I and II both agree in detail with Perm’s em-
pirical findings.

As the drive ( D ) increases from zero to D':

III. The reaction-evocation potential increases with a slight
positive acceleration.

IV. The reaction- evocation potential maintains its positive
growth relationship to the number of reinforcements. Both of these
corollaries agree in detail with Perm’s empirical findings.

As the drive ( D ) increases above D

V. There is a definite inflection in the 8 E R function at D, the
slope for values of D just greater than D' being less than for those
just below.

VI. The reaction-evocation potential above D' increases at first
with a slight positive acceleration, which soon gives place to a prac-
tically linear relationship. Both of these corollaries agree in detail
with Skinner’s empirical findings (Figure 54).

MISCELLANEOUS COROLLARIES FLOWING FROM THE PRESENT

PRIMARY MOTIVATION HYPOTHESIS

The first problem in this series is that presented by Elliott’s
experiment described above, the outcome of which is clearly shown
in Figure 53. Here we have the case of a reaction tendency set up
on the basis of one drive, showing a partial but by no means com-
plete disruption when this drive (thirst) is abruptly replaced by
another drive, that of hunger. At this point we recall the assump-
tion stated earlier (p. 241) that all drives alike are able to sensitize
all habits. Applying this to the behavior of Elliott’s animals dur-
ing the first critical test trial on the maze after the change in
drive, it is to be expected that while hunger was then the dominant

PRINCIPLES OF BEHAVIOR

248

drive, certain residual amounts of various other drives (including
thirst) were also active. These in the aggregate ( D), the hunger
drive included, are presumed to operate in a multiplicative manner
upon the habit strength effective at the moment in determining
reaction potentiality. It is assumed that this would be enough to
evoke, on the average, about 20 per cent as much activity as is
evoked by the thirst drive.

This means that the residual drive (Z)) must amount to con-
siderably more than 20 per cent of the regular thirst drive, say.
For example in the detailed analysis of the preceding section, where
24 hours’ hunger stood at 20 units of drive, this residual drive was
placed at 10 units, which is 50 per cent as much as 20. Neverthe-
less, the reaction potential at 24 hours’ hunger came out at 19.42
units, whereas that at satiation or zero drive stood at 4.58 units,
the latter being only about 23 per cent of the former. The ex-
planation of the paradoxical difference of 50 versus 23 per cent is
significant; it arises largely from the fact that when the hitherto
dominant drive ceases to be active, not only are there lost the
20 units of drive strength previously contributed by this need, but
there is also lost to the conditioned stimulus compound the sizable
component made up by S D , the withdrawal of which materially
reduces the available habit strength associated with the situation in
question, and so reduces the resulting reaction potential.

On the basis of the above analysis we may formulate the fol-
lowing additions to the corollaries listed in the preceding section:

VII. Under the conditions of the satiation of the dominant drive
involved in the original habit-acquisition, there are sufficient re-
siduals of other drives which in the aggregate yield on the average
an excitatory potential amounting to around 20 per cent of that
mobilized by a 24-hour hunger on a habit originally set up on the
basis of this drive.

VIII. In case an organism is presented with all the stimuli char-
acteristic of a habit, if the original drive is replaced by a strong
second drive whose S D activates no conflicting habit tendency, the
reaction potential to the execution of the habitual act will be
stronger than would be the case if the irrelevant second drive were
not active. This means that if a control group with both hunger
and thirst thoroughly satiated were to be added to the Elliott ex-
periment described above, the mean retardation in the running time
and the mean number of errors would increase appreciably above
what resulted from a mere replacement of one drive by another (6).

PRIMARY MOTIVATION AND REACTION POTENTIAL 249

A second problem concerns the relation of the experimental ex-
tinction of a reaction tendency to the drive intensity operative at
the time of extinction. Now, the passage from Pavlov quoted above
(p. 233) strongly suggests that experimental extinction effects are
in some sense directly opposed to reaction potential rather than
merely to habit strength. Since with a constant habit strength an
increase in the drive augments the reaction potential, and since
extinction effects are an increasing function of the number of unre-
inforced evocations, it follows that:

IX. The number of reinforcements being constant , the stronger
the relevant drive, the greater will be the number of unreinforced
evocations which will be required to reduce the reaction potential
to a given level.

X. The number of reinforcements being constant, the stronger
an allied but irrelevant drive active at the time of extinction, the
greater will be the number of unreinforced evocations required to
reduce the reaction potential to a given level, though this number
will be materially less than would be required under the same inten-
sity of the relevant drive .

Thus if a habit set up on the basis of a thirst drive were extin-
guished under a sizable hunger drive but with water satiation, the
theory demands that the reaction potential would extinguish with
fewer unreinforced evocations than would be the case under the
same intensity of the thirst drive in conjunction with a zero hunger
drive; moreover, such a habit would require more unreinforced
reaction evocations to produce a given degree of extinction under
a strong hunger drive than under a weak one. By the same type
of reasoning it is to be expected that if a reaction tendency were
set up in male rats under hunger or thirst, and if subsequently a
random sample of the organisms were castrated, experimental ex-
tinction under a normal hunger drive would occur more quickly
than it would in the non-castrated organisms.

At this point we turn to a more detailed consideration of Pav-
lov’s observation just referred to, that when he had performed an
experimental extinction under a given drive and then increased
the drive, the conditioned stimulus would again evoke the reaction.
This may be deduced rather simply: If a certain number of unrein-
forced evocations of a reaction have produced sufficient extinction
effects to neutralize a given amount of excitatory potential, an
increase in the drive will increase the excitatory potential which
the existent extinction effects will no longer suffice to neutralize

250

PRINCIPLES OF BEHAVIOR

completely. The balance of the reaction potential will accordingly
be available to evoke reaction and, upon adequate stimulation, will
do so. We thus come to our eleventh corollary:

XI. If a reaction tendency is extinguished by massed reaction
evocations under a given strength of drive, and if at once there-
after the drive is appreciably increased, the original stimulation
will again evoke the reaction.

Our final question concerns an exceedingly important problem
in adaptive dynamics. It has already been pointed out that as a
rule action sequences required to satisfy a food need are different
from those required to satisfy a water need, and both would ordi-
narily be quite different from the acts which would be required to
satisfy a sex drive. This problem is posed very sharply when, as
in the Hull-Leeper experiments, an organism is presented with an
identical objective situation and required to make a differential
reaction purely on the basis of the need dominant at the moment.
These experiments confirm everyday observations that animals can
adapt successfully to such situations. The question before us is
how this behavior is to be explained.

At first sight it might be supposed that in this situation the
animals would merely associate Sh with turning to the right, say,
and St with turning to the left, and that adaptation would thereby
be complete. A little further reflection will show, however, that
this simple explanation is hardly adequate, because if there were
really an independent and functionally potent receptor-effector con-
nection between the hunger-drive stimulus and turning to the right
the animal would, when hungry, be impelled to turn to the right
continuously when in its cage or wherever it happened to be, as well
as at the choice point in the maze. The animals, of course, display
no such behavior, any more than we ourselves do.

The present set of postulates mediates the explanation chiefly
on the basis of a secondary process known as patterning. Unfor-
tunately it will not be possible to give an exposition of this exceed-
ingly important subject until a later chapter (see p. 349 ff.). How-
ever, pending the detailed presentation in that place we shall here
merely indicate dogmatically the nature of patterning and briefly
sketch the application of this secondary pr in ciple to the problem in
adaptive dynamics now before us.

By the term “patterning” we mean the process whereby organ-
isms acquire the capacity of reacting (or not reacting) to partic-
ular combinations of stimuli as distinguished from the several com-

PRIMARY MOTIVATION AND REACTION POTENTIAL 251

ponent stimulus elements or aggregates making up the compound.
At bottom this process turns out to be a case of learning to dis-
criminate afferent interaction effects (p. 42 ff.). Specifically, the
principle of afferent interaction implies that in the Hull-Leeper
studies afferent impulses (s) arising from the environmental stimuli
(Si) are somewhat different when stimulation occurs in combina-
tion with the hunger-drive stimulus (5*) from those which result
from the same stimulation in combination with the thirst-drive
stimulus {St). Similarly, the afferent impulses arising from Sh and
St are somewhat different when initiated in conjunction with Sj
from those initiated by Sh and S t in the cage or other situations. If
the afferent impulses arising from the environmental stimuli un-
complicated by any particular drive be represented by s, then these
impulses when modified by the interaction with the hunger-drive
stimulus may be represented by sk, and when modified by interac-
tion with the thirst-drive stimulus, by s t . Since there are but two
alternatives, it is to be expected that at the outset of training, reac-
tion would be about 50 per cent correct. However, as the differen-
tial reinforcement yielded by the techniques employed in these
investigations continues, the gradient of generalization between s*
and st would progressively steepen, as shown in Figure 60; i.e.,
discrimination learning would gradually take place, exactly as it
does in fact. Thus we arrive at our twelfth corollary:

XII. Organisms will learn to react differentially to a given ob-
jective situation according to the drive active at the time, and to
react differentially to a given drive according to the objective situa-
tion at the time.

SUMMARY

The needs of organisms operate both in the formation of habits
and in their subsequent functioning, i.e., in primary motivation.
Because of the sensitizing or energizing action of needs in this latter
role, they are called drives.

A great mass of significant empirical evidence concerning pri-
mary motivation has become available within recent years. A sur-
vey of this material, particularly as related to hunger, thirst, injury
(including the action of very intense stimuli of all kinds), sex, and
the action of certain substances such as caffeine, has led to the
tentative conclusion that all primary drives produce their effects
by the action of various chemicals in the blood. Substances like
caffeine, through bathing the neural mechanisms involved, seem to

252

PRINCIPLES OF BEHAVIOR

operate by heightening the reaction potential mediated by all posi-
tive habit tendencies. Drive substances, such as the various endo-
crine secretions, are conceived either to be released into the blood
by certain kinds of strong stimulation or as themselves initiating
stimulation of resident receptors through their evocation of action
by selected portions of the body, e.g., the intestinal tract and the
genitalia. In both cases the energy effecting this receptor activa-
tion is called the drive stimulus ( S D ).

The action of these endocrine substances, while apparently low-
ering the reaction threshold of certain restricted effectors (1, p.
184 ff.), seems also to have a generalized but possibly weaker ten-
dency to facilitate action of all effectors, giving rise to a degree of
undifferentiated motivation analogous to the Freudian libido. Thus
a sex hormone would tend to motivate action based on any habit,
however remote the action from that involved in actual copulation.
This, together with the assumption that one or more other motiva-
tions are active to some degree, explains the continued but limited
amount of habitual action of organisms when the motivation on the
basis of which the habit was originally set up has presumably be-
come zero. It also suggests a possible mechanism underlying the
Freudian concept of sublimation. However, where differential be-
havior is required to bring about reduction in two or more drives,
the differences in the drive stimuli characteristic of the motivations
in question, through the principle of afferent interaction and the
resulting stimulus patterning, suffice to mediate the necessary dis-
crimination.

The hypothesis of the endocrine or chemical motivational mech-
anism and the associated principle of the drive stimulus, when
coupled with various other postulates of the present system such
as primary reinforcement, primary stimulus generalization, and
the opposition of experimental extinction to excitatory potential,
seem to be able to mediate the deduction, and so the explanation,
of nearly all the major known phenomena of primary motivation. 1
In addition to the phenomena already summarized there may be
mentioned the further deductions flowing from the system r that
resistance to extinction maintains a consistent growth function of
the number of reinforcements for any constant drive; that the

1 One class of phenomena seems to involve the action of fractional ante-
dating goal reactions and of spatial orientation. Space is not here available
for the elaboration of these mechanisms and their action in motivational
situations.

PRIMARY MOTIVATION AND REACTION POTENTIAL 253

asymptotes of these growth functions are themselves functions of
the strength of drive; that for constant habit strengths, reaction
potential has a positive acceleration for increasing drives between
zero and the drive employed in the original reinforcement; that if
habit strength is zero, reaction tendency is zero; that an increase
in drive will over-ride the total extinction of a reaction potential
arising from a weaker drive; that in a given objective habit situ-
ation the abrupt shift from one drive to another will, in the absence
of discriminatory training, disrupt the behavior to some extent,
though not completely; that transfers of training (habits) from one
motivation to another will be prompt and extensive ; that organisms
in the same external situations will learn to react differentially in
such a way as to reduce different needs; that the conditioned evoca-
tion of endocrine secretions facilitates the evocation of muscular
activity on the subsequent presentation of appropriate conditioned
stimuli, which is believed to be the role of “emotion” in the moti-
vation of behavior.

On the basis of the various background considerations elabo-
rated in the preceding pages, we formulate our sixth and seventh
primary molar laws of behavior:

POSTULATE 6

Associated with every drive ( D ) is a characteristic drive stimulus ( Sd )
whose intensity is an increasing monotonic function of the drive in
question.

POSTULATE 7

Any effective habit strength ( sHr ) is sensitized into reaction po-
tentiality ( sEr ) by all primary drives active within an organism at a given
time, the magnitude of this potentiality being a product obtained by
multiplying an increasing function of sHr by an increasing function of D.

From Postulates 5, 6, and 7 there may be derived the following
corollary:

MAJOR COROLLARY U

The amount of reaction potentiality (sEr) in any given primary motiva-
tional situation is the product of (1) the effective habit strength (s 1 + SqHr)
under the existing conditions of primary drive multiplied by (2) the
quotient obtained from dividing the sum of the dominant value of the
primary drive ( D ) plus the aggregate strength of all the non-dominant
primary drives ( D ) active at the time, by the sum of the same non-
dominant drives plus the physiological drive maximum (Md)»

>54

PRINCIPLES OF BEHAVIOR

where

NOTES

Mathematical Statement of Postulate 6
S D = 6/(D),
b > 0.

Mathematical Statement of Postulate 7

sE r =f( s H R ) Xf(D) (33)

Mathematical Statement of Major Corollary II

bE r = a x + s d H r 7~~Tr“» G**)

U + Md

where

D — the strength of the dominant primary drive at a moment under
consideration

D — the aggregate strength of all the non-dominant primary drives and
quasi-drives at the moment under consideration

Md — the physiological drive maximum (100 motes)

Si = the non-drive component of the Stimulus complex at the moment
under consideration

Sd => the stimulus specifically dependent upon the primary drive at the
moment under consideration

Bi+a D H H = the physiological summation of Bj H r and b d Hr

a i^ R ~ *h e effective habit loading of the non-drive component of the stimulus
complex

b d Hr = the effective habit loading of s D H R

The Equations of Perin’s Graphs

The curve drawn through the upper set of data points of Figure 50 was plotted
from the fitted equation :

n = 66(1 — 10 - 0180 at) __

where n represents the number of unreinforced reaction evocations to produce
experimental extinction and N represents the number of reinforcements in the
setting up of the habit. The curve drawn through the lower set of data points
of Figure 50 was plotted from the fitted equation :

n = 25(1 - 10- 0185 if) _ 4f

where n and N have the same significance as in the preceding equation. Note
the practical identity of the exponents, .0180 and .0185.

The curve drawn through the data points of Figure 49 was plotted from the
equation r

n = 9.4(10 0241 *) - 4,

PRIMARY MOTIVATION AND REACTION POTENTIAL 255

where n means the same as above and h represents the number of hours of food
privation.

The surface passing among the data points of Figure 51 was generated by
the fitted equation :

n = 21.45 (10 0222*)(i _ 10 - oi80 *) _ 4

(35)

in which n, and N mean the same as before. A comparison of this equation
with the preceding equations shows that it is essentially the positive growth
function of the first two equations in which the asymptote has been taken by
a function of the drive ( h ) derived from the third equation. Thus n, regarded
as action potentiality, may be seen to be a multiplicative function of h, or moti-
vation, and N, or habit strength.

The Equations Employed in the Derivation of Table 5 and Figure 56

The positive growth function from which the values in column 2 of Table 5
were derived is:

s 1 + s D H R = 75(1 - It)- ™ *).

The equation by which the values of column 3 were derived from those in
column 2 is:

aH B = 100 - V 10,000 - 100 sj + s d H r .

This equation is a special form of that representing the physiological summation
of two habit tendencies given below.

The equation from which the values of the drive (D) were calculated from
the number of hours' food privation ( h ) is:

100

120

The values of the drive deviations (d r ) were calculated from the equation :

d! = D' - D,

where D' represents the strength of drive employed in the formation of the habit
and D represents the strength of drive under which stimulation calculated to
lead to reaction evocation occurs.

The equation by means of which the values of columns 4, 5, 6 , 7, and 8 of
Table 5 were calculated from those of column 3 is :

S D H R = 8 D H R ( 10-00881 d0 .

This, it may be noted, is equation 29 (p. 199 ff.), the equation of primary stimulus
generalization in which b d Hr represents the effective-habit-strength loading of
the drive stimulus.

The equation by means of which the values of columns 9, 10, 11, 12, and 13 of
Table 5 were calculated from those of columns 4, 5, 6 , 7, and 8 is :

_ t7 1 r7 b^H r X s d H r
«!+ b d H r = Sjiift + b d H R —

The values of the effective drive (D) were found by the equation:

D = 100

£>+ D

D + Md

PRINCIPLES OF BEHAVIOR

in which D is supposed to be the sum of the generalized effects of all the irrelevant
drives active at the time, and Md represents the maximum drive possible in a
centigrade system, i.e., 100. The value of D found by trial to fit the Perm data
fairly well is 10. Therefore the equation becomes :

100

10 + D
110 *

For example, in case D is maximal this equation becomes:

100

110

= 100 .

The basic equation by means of which the values of columns 14, 15, 16, 17,
and 18 of Table 5 were calculated from the values of columns 9, 10, 11, 12, and 13

of Table 5, is :

bEr = Sj + b d Hr x

100 '

Substituting the equivalent of D and simplifying, this becomes:

bEr = b 1 + s d Hr X 1 9 1 ^ — .

REFERENCES

1. Bejach, F. A. Arousal, maintenance, and manifestation of sexual excite-

ment in male animals. Psychosomatic Medicine, 1942, 4, 173-198.

2. Elliott, M. H. The effect of change of drive on maze performance.

Calif. Pub. in Psychol., 1929, 4, 185-188.

3. Finch, G. Hunger as a determinant of conditional and unconditional

salivary response magnitude. Amer. J. Physiol., 1938, 123, 379-382.

4. Heathers, G. L., and Arakelian, P. The relation between strength of

drive and rate of extinction of a bar-pressing reaction in the rat. J •
Gen. Psychol., 1941, 24, 243-258.

5. Hovland, C. I. The generalization of conditioned responses: H. The sen-

sory generalization of conditioned responses with varying intensities of
tone. J. Genet. Psychol., 1937, 61, 279-291.

6. Hull, C. L. Differential habituation to internal stimuli in the albino rat.

/. Comp. Psychol., 1933, 16, 255-273.

7. Hull, C. L. The rat’s speed-of-locomotion gradient in the approach to

food. J. Comp. Psychol ^ 1934, 17, 393-422.

8. Leeprr, R. The role of motivation in learning: A study of the phenom-

enon of differential motivational control of the utilization of habits.
J. Genet. Psychol., 1935, 46, 3-40.

9. Miller, N. E., and Miles, W. R. Effect of caffeine on the running speed

of hungry, satiated, and frustrated rats. J. Comp. Psychol., 1935,
20, 397-412.

10. Moore, C. R., Price, D., and Gallagher, T. F. Rat prostate cystology

and testes-hormone indicator and the prevention of castration changes
by testes extract injection. Amer. J. Anat., 1930, 46, 71-108.

11. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

PRIMARY MOTIVATION AND REACTION POTENTIAL 257

12. Perin, C. T. Behavior potentiality as a joint function of the amount

of training and the degree of hunger at the time of extinction. J.
Exper. Psychol., 1942, SO, 93-113.

13. Skinner, B. F. The behavior of organisms. New York: D. Appleton-

Century Co., Inc., 1938.

14. Skinner, B. F., and Heron, W. T. Effect of caffeine and benzedrine

upon conditioning and extinction. Psychol. Record, 1937, 1, 340-346.

15. Stone, C. P. The retention of copulatory ability in male rats after

castration. J. Comp. Psychol ., 1927, 7, 369-387.

16. Stone, C. P. The retention of copulatory activity in male rabbits follow-

ing castration. J. Genet. Psychol., 1932, Ifi, 296-305.

17. Stone, C. P. Activation of impotent male rats by injections of testos-

terone proprionate. J. Comp. Psychol., 19S8, 25, 445-450.

18. Switzer, S. A. The effect of caffeine on experimental extinction of con-

ditioned reactions. J. Gen. Psychol., 1935, 12, 78-94.

19. Warden, C. J., Jenkins, T. N., and Warner, L. H. Introduction to com-

parative psychology. New York: Ronald, 1934.

20. Williams, S. B. Resistance to extinction as a function of the number

of reinforcements. J. Exper. Psychol., 1938, 23, 506-521.

21. Young, P. T. Motivation of behavior. New York: John Wiley and Sons,

1936.

22. Zener, K. E., and McCurdy, H. G. Analysis of motivational factors in

conditioned behavior: I. The differential effeet of changes in hunger
upon conditioned, unconditioned, and spontaneous salivary secretion.
J. Psychol., 1939, 8, 321-350.

CHAPTER XV

Unadaptive Habits and Experimental Extinction

Science tacitly assumes that similar causes will be followed by
similar effects. In the field of behavior dynamics it is accordingly
to be expected that an act which has been followed by a need
reduction in a given situation will always be so reinforced when-
ever the reaction occurs in other exactly similar situations. It may
be noted in this connection first that, as a rule, most of the factors
of such situations possess energies which activate one or more of
the receptors of the reacting organism. Secondly, according to the
law of reinforcement (p. 80 ff.), all stimuli whose receptor dis-
charges are contiguous with reactions which are followed by rein-
forcing states of affairs tend to acquire the capacity of later evok-
ing that reaction. As a result of this combination of circumstances
it comes about that on the recurrence of the situation in question
(including the need) the corresponding stimuli must also recur, they
will evoke the reaction conditioned to them, the need will be re-
duced, and survival will be facilitated.

At this point of the analysis, however, serious complications
appear. In the first place, exact duplicates of situations probably
never recur. A second complication is that by no means all of the
factors of any reaction situation are critical in the sense that their
presence is necessary for the act in question to bring about need
reduction. A third and closely related complication lies in the fact
that the really critical factor or factors of the reaction situation
may not stimulate the receptors of the reacting organism at all;
in the field of vision, for example, the view of the critical factor
may be cut off by the interposition of a completely irrelevant ob-
ject. Since organisms have no inner monitor or entelechy to tell
them m advance which stimulus elements or aggregates are asso-
ciated with the critical causal factor or factors of reaction situa-
tions, the law of reinforcement , other things equal , will mediate
the connections of the non-critical stimulus elements to the reaction
quite as readily as those of the critical ones.

As a result of the largely random flux of events in the world
to which organisms must react, it inevitably comes about that they
will often be stimulated by extensive groups of conditioned stim-

258

UN ADAPTIVE HABITS— EXPERIMENT AL EXTINCTION 259

ulus elements, none of which is causally related to the critical factor
or factors in the reinforcement situation. In such cases, if the
stimuli evoke the reaction it will not be followed by reinforcement.
This, of course, is wasteful of energy and therefore unadaptive.
The necessarily unadaptive nature of an appreciable portion of
the habits set up by virtue of the law of reinforcement naturally
raises the question of how organisms are able to survive under such
conditions (7, p. 501). The answer is found in the behavioral
principle known as experimental extinction. This is the subject
of the present chapter.

CONCRETE EXAMPLES OF EXPERIMENTAL EXTINCTION

The principle of experimental extinction is so ubiquitous a
factor in behavior dynamics that it can hardly escape anyone’s
observation. The dog which has been taught to “speak” for food
soon ceases to do this if the food, petting, etc., is systematically
withheld following the act. The principle has even passed into
folklore, as shown by the fable of the boy who shouted “Wolf!
Wolf 1” when no wolf was near. After a few such false alarms the
rescuing behavior of the hearers would inevitably become extin-
guished, and they would cease to respond to the calls quite as the
fable states.

The systematic investigation of experimental extinction origi-
nated in Pavlov’s conditioned-reflex laboratory in Petrograd. Since
the comparative simplicity of the conditioned-reflex technique
brings out with maximum clarity the essential principles involved,
a description of one of Pavlov’s experiments will serve as a useful
introduction to the technical aspects of the subject, even though
the artificiality of the experiment may tend somewhat to obscure
the functional significance of the principle. Pavlov reports having
produced a conditioned reflex by first showing a dog some meat
powder and then letting him eat it. After a considerable number
of reinforcements extending over several days, the mere visual stim-
ulation of the food would evoke a profuse flow of saliva. The meat
powder was then presented at a distance for a number of 30-second
periods, but without being followed by the customary feeding. On
each of the latter occasions the number of cubic centimeters of
saliva secreted was recorded. The results of this procedure are
shown in Table 6, where it may be seen (1) that after only a few
non-reinforced reactions the visual stimulus completely loses its

26o

PRINCIPLES OF BEHAVIOR

TABLE 6

Table Summarizing the Results of an Experiment Involving Expert
mental Extinction, Performed in Pavlov’s Laboratory ( 9 , p. 58).

Successive

Unreinforced

Stimulations

Number of Cubic Centimeters
of Saliva Secreted in

Each 30-Second Period

1.0

power of reaction evocation, and (2) that the course of this loss
is progressive, the rate of fall being more rapid at first than later.

EXPERIMENTAL EXTINCTION AS A FUNCTION OF THE NUMBER

OF UNREINFORCED REACTIONS

The process of experimental extinction has now been studied
in many other laboratories where many different reactions have
been extinguished under many different conditions. One of the
more easily interpreted of these studies has been reported by Hov-
land ( 6 ). This investigator conditioned a simple sinusoidal sound
wave of 1000 cycles per second to the galvanic skin reaction in
20 human subjects; the 24 reinforcements (a weak electric shock
to the wrist) by which the habit was originally set up were sepa-
rated by 30-minute rest pauses after the first and second series of
eight. The habit was then extinguished by repeated evocations of
the act without reinforcement. The pooled results of the first five
reactions of this extinction process presumably yield a set of values
closely approximating the typical rate of extinction. They are
represented graphically by the circles of Figure 57. In order to
determine more precisely the characteristics of this negative learn-
ing curve, a simple negative growth function was fitted to the
values represented by the circles. From this the smooth curve
passing among the circles was plotted. A glance at this curve shows
that, except at the zero point, the fit is excellent. The failure of
the amplitude of the reaction at point 0 to be as high, relatively,
as the other amplitudes may plausibly be interpreted as due to
“inhibition of reinforcement” (see p. 289 ff.), which is subsequently

UN ADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 261

“disinhibited” in part, at least, by the abrupt change in experi-
mental routine incidental to the process of extinction (5). We
conclude, then, that the curve of experimental extinction when un-
complicated by irrelevant factors is probably a simple negative
growth function.

Further examination of the smooth curve in Figure 57 reveals
two additional characteristics which merit consideration. The first

Fig. 57 . A graphic representation of the course of habit decrement as a
function of successive unreinforced evocations of the previously learned act.
(The data were received from Hovland in a private communication.)

is that the asymptote (limit of fall) of the curve is not zero, but
about 24 per cent of the amplitude of what the reaction was before
extinction began. This is probably an artifact due to the well-
known tendency of the skin to yield appreciable galvanic reactions
even to mild stimulations previous to any specific conditioning
whatever (see p. 186 ff.). In this respect it is believed that the
Pavlovian results shown in Table 6 more truly represent the quan-
titative principle of experimental extinction. The second notable

262

PRINCIPLES OF BEHAVIOR

characteristic of the function shown in Figure 57 is its strikingly
rapid rate of decrement, its F-value being a little less than By
way of contrast we may consider the rate of acquisition of a com-
parable habit, which is shown above in Figure 21 (p. 103). The
fractional increment (F) of this latter growth function is approxi-
mately 1 /14, which corresponds to a radically slower rate of change.
Thus we arrive at the indication that the rate of decrement under
the present set of conditions is more rapid than is the rate of the
original acquisition of the habit.

In this connection it must be pointed out that of those reactions
extensively investigated, the galvanic skin reaction and the salivary
reaction both show a progressive decrement in amplitude as experi-
mental extinction progressively reduces the excitatory tendency to
zero. The typical motor reaction differs sharply from this by dis-
playing under comparable conditions mainly an increase in latency
and a decrease in probability of occurrence. Motor reactions under
certain conditions, at least, are apt to show an increase in reaction
intensity in the early stages of extinction, though in the later stages
there is usually a slight tendency to a diminution in the intensity
of the reaction ( 2 , p. 148). Because of these differences it is prob-
able that the salivary reaction is the one best adapted for the
quantitative determination of the characteristics of the curves of
both simple learning and simple extinction.

THE STIMULUS GENERALIZATION" OF EXTINCTION EFFECTS

In an earlier chapter (p. 183 ff.) we saw that habits manifest
the phenomenon of stimulus generalization. Since physiologically
maladaptive habits (when first formed) are no different than other
excitatory tendencies, it is to be expected that they likewise will
generalize. We have also seen that experimental extinction pur-
ports to be a kind of corrective mechanism. It is evident, however,
that if experimental extinction is to correct false habit tendencies
with reasonable efficiency, extinction effects must also generalize;
if extinction were a mere point phenomenon, false receptor-effector
connections would probably never become wholly eradicated be-
cause the process of extinguishing a zone consisting of infinitesimal

points would be literally interminable.

As a matter of fact, extinction effects do manifest stimulus

generalization to a marked degree. This important phenomenon.

UNADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 263

like so many others in this field, also appears first to have been
investigated in Pavlov’s laboratory. For example, the sound of
a buzzer, the sound of a metronome, and a tactile stimulation were
separately conditioned in a dog to the salivary reaction induced
by having weak acid injected into his mouth. At the conclusion
of the positive training the buzzer evoked 13 drops per 30-second
period, the metronome, 12 drops, and the tactile stimulus, 4 drops.
The metronome was then extinguished by unreinforced presenta-
tions at 3-minute intervals until it evoked no secretion whatever.
A few minutes later the secretion evoked by the tactile stimulus
had fallen to zero and that by the buzzer had fallen to 2.5 drops
( 9 , p. 55). The effects of the extinction of the metronome habit
had clearly generalized in such a way as to inhibit completely the
tactile habit and almost to inhibit the buzzer habit.

Pavlov even investigated the quantitative gradient of general-
ized extinction effects. He set up a number of “homogeneous”
conditioned reflexes to a series of points on the skin, then extin-
guished the reaction whose stimulus was located at one extreme of
the series and noted the extent to which the other conditioned
reflexes were weakened as a result. On the basis of such experi-
ments, Pavlov concludes ( 9 , p. 158) :

It is plain that, the further away on the skin the secondarily inhibited
place is from the place which undergoes the primary inhibition [extinc-
tion], the weaker is the irradiated inhibitory after-effect.

Subsequently, Anrep ( 1 ), Bass and Hull ( 8 ), and Ho viand ( 6 )
sought to plot the generalization gradient of extinction effects with
progressively more refined experimental procedures. Anrep em-
ployed cutaneous stimulation with the salivary reaction to food
in dogs; Bass and Hull employed cutaneous stimulation with the
galvanic skin reaction evoked by an electric shock in humans.

Hovland conditioned the pitch of four pure tones an equal
amount to the galvanic skin reaction evoked originally by a weak
electric shock. The tones were so chosen that they differed from
each other by 25 discrimination thresholds (j.n.d.’s). Then a tone
at one or the other extreme of the series was extinguished to a
partial but known degree, after which all four tones were tested to
determine the strength of the residual excitatory tendency evocable
by each. The pooled results of the twenty subjects employed in
this investigation are represented graphically by the circles in

PRINCIPLES OF BEHAVIOR

264

Figure 58. This series of circles constitutes a clear verification of
the gradient of generalized extinction effects reported by Pavlov.
It is evident that experimental extinction is an extended and not
a point phenomenon.

The precision and general reliability of Hovland’s data also
warranted an attempt at a determination of the mathematical

d IN j.n.d. UNITS

Fia. 58. Empirical stimulus generalization gradient of an extinguished
galvanic skin reaction plotted from data published by Hovland ( 6 ). The
gradient extends in both directions on the stimulus continuum (rate of
vibration) from the point extinguished (0); d is the difference between the
stimulus originally conditioned and the stimulus evoking the reaction. Note

that here the gradients are directly the reverse of those shown in the closely
related Figure 42, p. 185.

characteristics of the gradient. To this end, an equation was fitted
to the values represented by the circles. From this equation was
plotted the smooth curve running among them. That the equation,
a simple positive growth function, fits the data rather well may
be seen by an inspection of the figure. We accordingly conclude
that the gradient of stimulus generalization extinction effects is
probably a simple positive growth function.

UNADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 265

THE INTERACTION OF THE GRADIENTS OF EXCITATION AND OF

EXTINCTION

It is clear from a comparison of Hovland’s excitation and ex-
tinction gradients (Figures 58 and 42, p. 264 and 185), that if their
parameters turn out to be alike the second is exactly the shape
which would be required completely to eliminate the first. The
conditions of the two experiments are such that an exact agree-
ment is not to be expected between the two maximum opposing
values. In the matter of the constant incremental factor of change

Fia. 59. Diagram representing the manner in which the gradients of ex-
perimental extinction are supposed, theoretically, to interact with the gradients
of a false or unadaptive reaction tendency in such a way as to eliminate the
latter. The excitatory gradients are represented by the upper curves, those
of extinction effects are represented by the lower curves, and the residue of
effective excitatory tendency is represented by the dotted line in between.

(F), which might easily be the same, we find that agreement also
does not exist; the F-f actor of the excitation generalization is ap-
proximately 1/33, whereas that of extinction is approximately 1/21,
the range of the former being appreciably the greater. A difference
of this order might, however, easily arise from sampling “errors.”
A considerable amount of careful quantitative research is badly
needed to clear up this matter.

The manner in which the two gradients are conceived to inter-
act in the elimination of the unadaptive effects of false (and there-
fore unadaptive) reaction tendencies is shown in Figure 59. In
this figure the upper pair of gradients represent excitation. They

266

PRINCIPLES OF BEHAVIOR

are drawn on the assumption that reinforcement occurred at point
zero on the stimulus continuum and diminished at the rate of
approximately 1/29 at each additional j.n.d. of deviation of the
evoking stimulus. This value was chosen because it falls about
midway between the E-values of Hovland’s two gradients. Extinc-
tion is assumed also to have taken place at zero on the stimulus
continuum, and to have continued until no reaction would be evoked
by that stimulus, i.e., until the reaction tendency had passed be-
neath the reaction threshold. Since the reaction threshold is taken
at 10 units, this means that the extinction effects must have become
great enough to neutralize 90 — 10, or 80 points of excitation; i.e.,
the extinction effects must possess a negative strength of 80 units.
The generalization gradients of these extinction effects are plotted
on the basis of the same E-constant as are those of the excitation
effects, viz., 1/29. By subtracting the extinction gradient from
the corresponding excitation gradient, we obtain the effective or
residual strength of the excitatory tendency, which is represented
by the dotted line. This shows that at all points on the stimulus
continuum the effective excitatory tendency is below the reaction
threshold, i.e., at no point on the stimulus continuum will the
unadaptive reaction be evoked.

A rather different picture, and one of great theoretical signifi-
cance, emerges when extinction takes place, not at the point of
reinforcement but out on one of the wings of the excitation gradi-
ent ( 8 , p. 25) . In nature this typically occurs when generalization
has extended on the stimulus continuum to a point where the reac-
tion will no longer be reinforced. Such a situation calls for dis-
crimination on the part of the organism; thus primitive stimulus
generalization is in a sense indiscriminate and is the natural anti-
thesis of discrimination. Por example, in Figure 59 the excitatory
tendency before extinction would evoke reaction with varying de-
grees of intensity or probability anywhere on the stimulus con-
tinuum within about 64 j.n.d. ’s of the point originally reinforced.

Suppose, however, that the stimulus at 8 j.n.d/s on one side
of the point of reinforcement represents a state of affairs which
in conjunction with the habituated act will not yield reinforcement.
Suppose further that this stimulus is presented to the subject re-
peatedly until it will no longer evoke the reaction to any degree
whatever. The interaction of the generalization gradients of the
extinction effects so produced, with those of the original excitation
gradients, is shown in Figure 60. There, as in Figure 59, it may

UN ADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 267

be seen by the course of the dotted line that the point extinguished
diminishes the effective excitatory tendency to the reaction thresh-
old, a reduction of some 58 points. It will also be noticed that at
no place beyond this point does the effective excitatory tendency
rise above the reaction threshold. However, between the point
extinguished and the point of original reinforcement the curve of
effective excitation rises steeply to a value of 45, which is far above
the reaction threshold. This means that through the interaction

80 60 40 20 0 20 40 60

DEVIATION Id) IN j.n.d.s FROM POINT OF REINFORCEMENT

Fia. 60. Diagram representing the manner in which the gradients of ex-
perimental extinction are supposed, theoretically, to interact with the gradients
of a positive reaction tendency to generate the phenomenon of discrimination
learning. As in Figure 59, the upper curve represents excitation, the lower
curves represent extinction effects, and the dotted line in between represents
the residual effective reaction tendency. Note the greatly steepened gradient
on the latter curve between 0 and 8 j.n.d/s. It is largely because of this that
the improvement in simple discrimination is believed to occur.

of the gradients of excitation and extinction the phenomenon of
simple discrimination learning has been generated as a secondary
principle . x

THE REACTION GENERALIZATION OF EXTINCTION EFFECTS

Just as the adaptive aspects of behavior dynamics require a
stimulus generalization of extinction effects to neutralize effectively

1 This view of simple discrimination learning follows substantially the ap-
proach developed by Spence in numerous theoretical and experimental studies

uo, 11, 12, is, 14).

268

PRINCIPLES OF BEHAVIOR

the unadaptive habits inevitably set up in considerable numbers
by the indiscriminate action of the law of reinforcement, so the
reaction generalization tendencies of unadaptive habits require for
the survival of organisms that there be a complementary reaction
generalization tendency of extinction effects. Here again Pavlov
made the initial discovery. He reports ( 9 , p. 54) :

This latter phenomenon [generalized extinction effects] involves not
only those conditioned reflexes which were based upon a common uncon-
ditioned reflex with the primarily extinguished one (homogeneous coris-
ditioned reflexes), but also those which were based upon a different un-
conditioned reflex (heterogeneous conditioned reflexes).

In Pavlov’s terminology, if a tone and a tactile stimulus were each
separately conditioned to an alimentary salivary secretion by the
use of food reinforcement, the resulting reflexes would be homo-
geneous. If, on the other
hand, the tone were rein-
forced with food and the
tactile stimulus were rein-
forced by weak acid being
injected into the mouth, the
tactile conditioned reaction
would be defensive and ob-
servably different in nature;
the latter two conditioned
reactions would accordingly
be called heterogeneous.

An analogous phenome-
non in a selective learning
situation was reported by
Youtz ( 15 ) ; a closely related
study worked out with me-
ticulous care has been per-

Fia. 61. Sketch showing the two manip- f ° rm , ed ^ EUs0n <4h
illation bare employed in Ellson’s experi- the latter investigation ai-

ment ( 4 , P* 341). bino rats were placed, one at

a time, in a small, sound-
shielded, cubical space. From one wall of the chamber there pro-
jected two bars, as shown in Figure 61. These bars could be in-
troduced into the chamber or retracted at will, without opening
the chamber door. During the learning and extinction processes

UNADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 269

only one bar was presented at a time. A lump of moist food was
placed behind the panel so that its odor would be carried through
the bar slot by the ventilating system. By investigating this odor
the animal sooner or later would move the bar in question, the
horizontal bar downward, and the vertical bar to the left. Either
movement caused a magnetic release mechanism behind the panel
to give a click and at the same time a cylinder of food dropped
into a cup beneath. In this way the horizontal-bar habit was set
up ; the vertical-bar habit was set up 30 minutes later in an exactly
analogous manner.

On the next day the two habits were extinguished in succession,
first the vertical-bar habit, and 5.5 minutes later the horizontal-
bar habit; this was done by severing the electrical connection be-
tween the manipulative bars and the food-release mechanism. The
horizontal-bar habit was extinguished in control animals without
a preceding extinction on the vertical-bar habit. It was found that
on the average the control animals operated the horizontal bar
49.7 times before a given degree of extinction supervened, whereas
5.5 minutes after the extinction of the vertical-bar habit the experi-
mental animals extinguished on the horizontal bar to the same
degree after only 23.9 operations of the bar. This reduction in
resistance to experimental extinction of approximately 50 per cent
clearly suggests reaction generalization of extinction effects.

It is to be noted, incidentally, that neither in the Pavlovian
experiments nor in that of Ellson is the reaction generalization
uncomplicated by stimulus generalization. In both cases there was
a decided element of stimulus similarity between the original ex-
tinction and the subsequent generalization-of-extinction situation
in that the incidental stimuli from the apparatus environment were
identical. Nevertheless, these experiments demonstrate that extinc-
tion effects are transferable from one reaction to another reaction
which is to a considerable extent different.

THE SPONTANEOUS RECOVERY OF EXTINCTION EFFECTS

The account of experimental extinction contained in the preced-
ing sections of the present chapter, together with the somewhat
misleading use of the word extinction in this connection, might
easily suggest to the uninitiated that experimental extinction neces-
sarily abolishes completely and permanently the unadaptive reac-
tion tendency involved. This is far from being the case, as Pavlov

270

PRINCIPLES OF BEHAVIOR

himself long ago pointed out ( 4 , p. 60). That a reaction tendency-
may be very much alive after total experimental extinction is
demonstrated in a striking manner by the fact that if the condi-
tioned stimulus is withheld from the organism for some time after
experimental extinction has occurred, its reapplication will evoke
the reaction to a considerably greater extent than it did at the
conclusion of the original extinction. This is known as spontaneous
recovery.

The phenomenon of spontaneous recovery was discovered by
Pavlov; there is accordingly a certain appropriateness in choosing
our initial illustration of it from his writings. This example comes
as a sequel to an extinction experiment reported above (p. 259 ff.)
and summarized in Table 6. Following the extinction there de-
scribed, the dog was left to itself for two hours, after which the
conditioned stimulus (visual presentation of meat powder) was
again delivered. This was followed by .15 cubic centimeters of
salivary secretion, which showed that the reaction tendency had
spontaneously recovered about one-sixth of its original strength.
Usually the rate of recovery is greater than this.

Proceeding now to the consideration of the quantitative law
of the spontaneous recovery of experimental extinction as a func-
tion of time, we turn to another portion of the Ellson experiment
described in the preceding section of the present chapter (p. 268 ff.).
Four groups of 25 albino rats were trained by means of an equal
number of food reinforcements to depress a horizontal bar (Fig-
ure 61) until the habit was thoroughly learned. They were then
presented with the bar and permitted to operate it without rein-
forcement until a period of five minutes elapsed without recorded
pressures. Following this one group was extinguished to the same
degree a second time after a recovery period of 5.5 minutes, another
group was again extinguished after 25 minutes, a third group after
65 minutes, and a fourth group after 185 minutes. The solid circles
in Figure 62 show the median number of unreinforced reactions
required again to produce extinction after the several recovery
periods. It is evident at a glance that spontaneous recovery is
considerable in amount, and that it is an increasing function of
the duration of the recovery period.

In an attempt to determine this relationship more precisely, a
positive growth function was fitted to these values; this is repre-
sented by the smooth curve drawn among the solid circles. While
the fit is by no means perfect, it is evident that the spontaneous

UNADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 271

recovery from primary experimental extinction effects is approxi-
mately a simple positive growth function. An inspection of this
curve shows that at 185 minutes it practically reaches its asymp-
tote, or limit of rise. It is also to be noted that this maximum is
only about 50 per cent of the original strength of the habit, which
is indicated by the broken line at the top of the figure.

We saw in the last section (p. 267 ff.) that extinction effects
manifest reaction generalization, which raises the question of the

Fia. 62 . Graphic representation of Ellson’s empirical values for the
spontaneous recovery of a habit from primary experimental extinction (lower
curve, solid circles) and from reaction-generalized experimental extinction
effects (upper curve, hollow circles). Both curves represent simple positive
growth functions fitted to the circles through which they pass. (Plotted from
data published by Ellson, 4 .)

quantitative law regarding the spontaneous recovery of the reaction
generalization of extinction effects. This problem also was inves-
tigated by Ellson as a portion of the experiment just described.
Four additional groups of animals were extinguished on the ver-
tical-bar habit, and then after recovery periods of 5.5, 25, 65, and
185 minutes they were extinguished on the horizontal-bar habit.
The median number of unreinforced reactions required to produce
experimental extinction are shown by the hollow circles of Fig-
ure 62.

A simple growth function was fitted to these values; this is

PRINCIPLES OF BEHAVIOR

272

represented by the smooth, broken-line curve drawn among the
circles. Here again the fit is not very close, though there is an
indication that spontaneous recovery from reaction-generalized ex-
tinction effects approximates a simple growth function. It is to be
noted that this curve also approximately reaches its limit of rise
at 185 minutes, which, incidentally, almost exactly equals the
number of unreinforced reactions required to extinguish the hori-
zontal bar when this extinction is not preceded by the extinction of
the vertical-bar habit. Moreover, it is probably significant that
the rate of rise of this curve is very close to that of the one fitted
to the data derived from the spontaneous recovery of primary ex-
tinction effects; the fractional rate of change ( F ) of the latter is
1/21, and that of the former is approximately 1/24.

THE DISINHIBITION OF EXTINCTION EFFECTS

A second phenomenon which demonstrates that experimental
extinction to the point of zero reaction does not necessarily abolish
a reaction tendency permanently and completely is that known as
disinhibition. This was pointed out by Pavlov, in whose labora-
tory the phenomenon was originally discovered. The nature of
disinhibition is nicely illustrated by the following account.

Dr. Zavadsky, one of Pavlov’s pupils, experimented with a dog
which had two salivary fistulas, one from the submaxillary and
the other from the parotid gland. Through repeated presentations
and ingestions, the sight and odor of meat powder presented at
a distance had become conditioned to evoke the salivary reaction
of both glands. Thereupon occurred the events summarized in
Table 7, where it may be seen that the first three stimulations
were not reinforced. An extremely rapid experimental extinction
resulted, shown by the fact that at the third stimulation the reac-
tion was zero. On the fourth trial, however, the presentation of
the meat powder was accompanied by an “extra” stimulus in the
form of a cutaneous vibration. In this case three drops of saliva
were secreted, which indicates that the inhibition was partly abol-
ished (disinhibited) . Five minutes later when the meat powder
was presented it was accompanied by knocks under the table; dis-
inhibition was again manifested by a secretion of two drops. After
five minutes the meat powder was presented alone; the zero reac-
tion to this indicates that the extinctive inhibition had returned,
following the preceding disinhibition. This observation illustrates

UNADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 273

TABLE 7

Summary of Dr. Zavadsky’s Experimental Results Illustrating Both
Simultaneous and Perse verative Aspects of Disinhibition ( 9 , p. 65).*

Time of

Stimulus Applied During One Minute

Amount of Saliva in Drops
During One Minute

Occurrence

From Submaxil-
lary Gland

From Paro-
tid Gland

1:53 p.m.

Meat powder presented at a distance

. 7

1:58 p.m.

Meat powder presented at a distance

2:3 p.m.

Meat powder presented at a distance

2:8 p.m.

Same tactile stimulation of skin . . .

2:13 p.m.

Same + knocks under the table

2:18 p.m.

Meat powder at a distance

2:20 p.m.

Prof. Pavlov enters the room contain-
ing the dog, talks, and stays for two
minutes

• • •

2:23 p.m.

Meat powder at a distance

2:28 p.m.

Same

* Previous to this experiment it had been shown repeatedly that neither the tactile nor the
auditory stimulus, nor the entry of Professor Pavlov into the experimental room, produced any
secretory effect at alL

the transitory nature of disinhibition, which presumably is due to
the fading out of the stimulus trace of the extra stimulus. It is to
be noted (Table 7) that no disinhibition was followed by a reaction
nearly as great as the eleven drops evoked at 1:53 p.m. preceding
the experimental extinction.

The reactions just considered are cases of simultaneous disin-
hibition. In spite of its clearly transitory nature, disinhibition
shows a certain tendency to perseveration, or after-effect. The
entrance of Professor Pavlov into the experimental room for two
minutes served as an obvious external stimulation. One minute
after he had left the meat powder was presented at a distance and
it evoked the conditioned reaction of five drops; this illustrates the
perseverative effect of disinhibition. Five minutes later, however,
the same stimulus produced no reaction, which shows that the per-
severation was distinctly brief in duration.

SUMMARY

The conditions under which reinforcements occur inevitably set
up many receptor connections that are false in the sense that if
the stimulus elements in question should alone evoke the reaction,

274

PRINCIPLES OF BEHAVIOR

reinforcement would not follow. The functional corrective of this
unadaptive aspect of the law of reinforcement is experimental ex-
tinction. This consists of a progressive weakening of the reaction
tendency whenever the evocation of the reaction is not followed
by adequate reinforcement. Experiments suggest that the reaction
tendency diminishes as a negative growth function of the number
of closely successive unreinforced evocations. In one study in
which analysis was made of the relevant learning curves, it was
found that the rate of loss through extinction was much faster than
the rate of the original acquisition of the reaction tendency.

Just as positive habits manifest stimulus generalization, so do
extinction effects also manifest this tendency. Moreover, the gradi-
ent of the generalization of extinction effects is of such a nature
that if the incremental factor ( F ) in the two cases were the same
the extinction of a habit at the point of reinforcement on the
stimulus continuum would completely neutralize the reaction ten-
dency, not only at that point but at all other points on the stim-
ulus continuum to which it would show primary stimulus general-
ization.

In case experimental extinction occurs on one wing of a posi-
tive generalization gradient, the interaction of this with the result-
ing extinction generalization gradient produces a greatly steepened
gradient of that portion of the effective reaction tendency lying
between the point of extinction and the point of the original rein-
forcement. This steepened gradient leaves the latter stimulus still
able to evoke the reaction, while the former does not. The result
is a distinctly heightened power of discrimination.

In both conditioned-reflex and selective-learning situations, ex-
tinction effects manifest reaction generalization, quite as do posi-
tive habit tendencies.

Experimental extinction does not necessarily abolish completely
and permanently the reaction tendency extinguished; this is shown
by the phenomenon of spontaneous recovery. Spontaneous recov-
ery, of both primary extinction and response-generalization extinc-
tion effects, takes place approximately as a positive growth function
of time elapsing since the termination of the extinction process.
Under the continuous action of a static conditioned stimulus, spon-
taneous recovery of both primary and generalized extinction nearly
reaches its maximum at three hours. Maximal spontaneous recov-
ery of primary extinction effects under these conditions is about

UN ADAPTIVE HABITS— EXPERIMENTAL EXTINCTION 275

50 per cent, whereas that of generalized extinction effects is ap-
proximately 100 per cent.

A second phenomenon, known as disinhibition, further demon-
strates that experimental extinction does not necessarily constitute
an abolition of the extinguished reaction tendency. This means
that a weak “extra” stimulus will partially restore an extinguished
reaction tendency. Such restorations are quite transitory, however.
Disinhibition may partially restore an extinguished reaction ten-
dency even when the extra stimulus has been delivered several min-
utes before the attempt is made to evoke the extinguished reaction.

NOTES

The Equation Fitted to Hovland’s Curve of Extinction Data

Following Distributed Practice

The equation from which was plotted the smooth curve of Figure 57 is :

A = 24.1 + 80.9 X 10 - - 315 ”,

where A is the amplitude of the galvanic skin reaction evoked by the conditioned
stimulus, and N is the number of the preceding extinction repetitions. In this
equation the exponential value of .315 corresponds to a factor of reduction ( F )
of 1/1.94.

The Equation Fitted to Hovland’s Generalized Extinction Data
The equation from which the smooth curve of Figure 58 was plotted is:

A = 6.7 + 3.25(1 - 10“ 0196 d ),

where A is the amplitude of the galvanic skin reaction evoked by a stimulus,
and d is the difference in j.n.d.’s between that stimulus and the one to which
the reaction was originally extinguished.

The comparable equation fitted to the same author’s published empirical data
on excitatory generalization effects is:

A = 12.6 + 6 X 10- 0135 d .

The F-value corresponding to .0135 is approximately 1/33; that corresponding
to .0195 is approximately 1/21. Unfortunately it is not known whether either
of these F-values is constant under all conditions, and if not, upon what their
magnitude may depend.

The Equations of the Curves Shown in Figure 62

The equation of the curve of the spontaneous recovery of primary extinction
effects shown in the lower portion of Figure 62 is:

n = 22.2 (1 - 10 - ow »"') _ 2 ,

where n is the number of unreinforced reactions required to produce the second
experimental extinction, and t'" is the time in minutes of the recovery period.

PRINCIPLES OF BEHAVIOR

276

i.e., from the termination of the first experimental extinction to the beginning
of the second experimental extinction.

The equation of the curve of spontaneous recovery from the reaction generali-
zation of experimental extinction is :

» = 43 - 32 X 10- 021

where n has the same significance as in the preceding equation, and t"' is the
time in minutes of the recovery period, i.e., from the termination of the extinction
of the vertical-bar habit to the begi nni ng of the extinction of the horizontal-bar
habit.

The rate of rise to its asymptote of the curve represented by the first equation
(F) is approximately 1/24; that of the second is 1/21.

REFERENCES

1. Anrep, G. V. The irradiation of conditioned reflexes. Proc. Royal Soc.

London, 1923, 94, Series B, 404, 425.

2. Arakelian, P. Cyclic oscillations in the extinction behavior of rats. J.

Gen. Psychol., 1939, 21, 137-162.

3. Bass, M. J., and Hull, C. L. The irradiation of a tactile conditioned

reflex in man. J. Comp. Psychol., 1934, 17, 47-65.

4. Ellson, D. G. Quantitative studies of the interaction of simple habits.

I. Recovery from specific and generalized effects of extinction. J. Exper.
Psychol., 1938, 23, 339-358.

5. Hovland, C. I. ‘Inhibition of reinforcement’ and phenomena of experi-

mental extinction. Proc. Natl. Acad. Sci., 1936, 22, 430-433.

6. Hovland, C. I. The generalization of conditioned responses: I. The sen-

sory generalization of conditioned responses with varying frequencies of
tone. J. Gen. Psychol., 1937, 17, 125-148.

7. Hull, C. L. A functional interpretation of the conditioned reflex. Psy-

chol. Rev^ 1929, 36, 498-511.

8. Hull, C. L. The problem of stimulus equivalence in behavior theory.

Psychol. Rev^ 1939, 49, 9-30.

9. Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

10. Spence, K. W. The nature of discrimination learning in animals.

Psychol. Rev^ 1936, 43, 427-449.

11. Spence, K. W. The differential response in animals to stimuli varying

within a single dimension. Psychol. Rev^ 1937, 44, 430-444.

12. Spence, K. W. Analysis of formation of visual discrimination habits in

chimpanzee. J. Comp. Psychol., 1937, 23, 77-100.

13. Spence, K, W. Failure of transposition in size discrimination of chim-

panzees. Amer. J. Psychol., 1941, 64, 223-229.

14. Spence, K. W. The basis of solution by chimpanzees of the intermedi-

ate size problem. J. Exper. Psychol., 1942, 31, 257-271.

15. Youtz, R. E. P. The weakening of one Thomdikian response following

the extinction of another. J. Exper . Psychol , 1939, 24, 294-304.

CHAPTER XVI

Inhibition and Effective Reaction Potential

In the last chapter we considered experimental extinction as
the mechanism which protects organisms from the evil effects of
the unadaptive habits inevitably set up by the law of reinforce-
ment. In the present chapter we shall again take experimental
extinction as our point of departure; here, however, it will be
regarded as a secondary phenomenon which arises under certain
special conditions from the logically more primitive principle of
reactive inhibition. Our expository procedure will be to state in a
semi-formal manner certain principles, more or less physiological
or submolar in nature, according to which reactive inhibition is
believed to originate, operate, and disintegrate, and to accompany
them with illustrative evidence. These submolar principles, it is
to be noted, are not properly a part of the present system, which
is molar, but are intended as a kind of background. Following this
there will be presented a series of corollaries flowing from these and
other principles of the system . 1 In this way an attempt will be
made to show how the principles explain, and therefore integrate,
an appreciable variety of relevant empirical phenomena. Finally
two primary molar principles will emerge and will be formally
stated as such at the end of the chapter.

QUANTITATIVE CONCEPTS AND PRELIMINARY STATEMENT OF
PRIMARY MOLAR AND SUBMOLAR PRINCIPLES RELATED

TO INHIBITORY POTENTIAL (/«)

Although the physiology of response inhibition is far from clearly
known, a great deal of knowledge of a submolar nature has been
discovered during the last quarter century. Our account of this
subject will therefore proceed at first with submolar principles as

1 For expository purposes the preliminary propositions and the corol-
laries which flow from them are more or less alternated in the following
pages. It will be noted that the preliminary propositions are indicated by
capital letters, whereas the corollaries are indicated, as in other chapters,
by Roman numerals.

277

PRINCIPLES OF BEHAVIOR

278

a background, the Mowrer-Miller hypothesis 1 being taken as a
point of departure. While this hypothesis has a number of com-
ponents which appear in various parts of the present chapter, the
main or critical proposition may be stated as our first preliminary
or submolar principle.

A. Whenever any reaction is evoked in an organism there is
left a condition or state which acts as a 'primary negative motiva-
tion in that it has an innate capacity to produce a cessation of the
activity which produced the state.

We shall call this state or condition reactive inhibition. From
a quantitative point of view, reactive inhibition will be repre-
sented by the symbol I R . Just as g E R symbolizes a certain quan-
tity of reaction evocation potential, so I R symbolizes a certain
potentiality of inhibition, i.e., a certain quantity of inhibitory
potential. The reaction decrement which we have attributed to
reactive inhibition obviously bears a striking resemblance to the
decrements w r hich are ordinarily attributed to “fatigue." It is
important to note that “fatigue" is to be understood in the present
context as denoting a decrement in action evocation potentiality,
rather than an exhaustion of the energy available to the reacting
organ {17).

From the foregoing it is evident that inhibitory potential ( Ir )
is an unobservable and so has the status of a logical construct with
all the advantages and disadvantages characteristic of such scien-
tific concepts. In this connection it will be recalled (p. 21 ff.)
that the prime prerequisite for the proper use of unobservables in
scientific theory is that they be anchored in a quantitatively unam-
biguous manner (a) to observable antecedent conditions or events,
and (6) to observable consequent conditions or events.

Proceeding at once to the satisfaction of the first of these
requirements 2 in so far as the present state of our ignorance per-

1 A brief statement of Dr. Mowrer’s version of the Mowrer-Miller hypoth-
esis is contained in an article by himself and Miss Jones (15) ; Dr. Miller’s
version is presented in his recently published book U4, p. 40 ff.). We are
much indebted to Dr. Mowrer not only for material appearing in the pres-
ent section, but for ideas scattered throughout the entire chapter; however,
for the particular formulation of the hypothesis here presented, in so far as
it differs from the views of Dr. Mowrer and Dr. Miller, as well as for the
deduction of the most of the corollaries derived in one way or another
from it, the present author takes entire responsibility.

2 The second requirement not only of the present unobservable but of
a number of others employed in earlier chapters will be taken up in Chap-
ter XVIII.

INHIBITION-EFFECTIVE REACTION POTENTIAL 279

mits, we arrive at our second preliminary or submolar proposi-
tion:

B. The net amount of functioning inhibitory 'potential resulting
from a sequence of reaction evocations is a positively accelerated
function of the amount of work (TV) involved in the performance
of the response in question.

Stated still more specifically, it may be said that the weight
of the evidence at present available indicates that

where n is the number of reaction evocations involved, c and B
are empirical constants, and

W = F'L,

in which F f represents force and L represents distance or length
of the movement, as in ordinary mechanics.

It is evident that the mean net increment of inhibition per rein-
forcement must be the net inhibition divided by the number of
reaction evocations. From this consideration and the basic equa-
tion for I, it follows that for a given organism the mean net incre-
ment of inhibition must have a constant value for a given value
of W, i.e., it must be

B — W

It is to be noted in this connection, however, that the inhibitory
potential resulting from a series of motor responses is not a simple
matter of mechanics, that it does not depend merely upon the
force (F') and distance (L) involved in the movement. This is
precluded by the constant, c, in the relationship. For example, it
is presumable that for a given amount of energy consumption such
as is required for the repeated lifting of a heavy weight, the value
of c would be larger for the weak muscular system involved in the
flexing of the little finger than for the relatively strong muscular
system which flexes the arm at the elbow.

The relationship of energy expenditure or work ( W ) to the
accumulated inhibition I R arising from a sequence of unreinforced
reaction evocations, such as occur in experimental extinction, is
convincingly demonstrated in an investigation reported by Mowrer
and Jones (16). Three comparable groups of albino rats were
trained on a Skinner type of apparatus to press a bar for food

28o

PRINCIPLES OF BEHAVIOR

pellets. The bar was so constructed that different weights could
be attached to it requiring the animal to make any desired min i-
mal pressure before the food would be delivered. The animals of

all three groups were given equal preliminary reinforcement with
bar weights of 5 grams, 42.5 grams, and 80 grams. Following this
the three groups were extinguished during three periods of 20 min-
utes each, with 24 hours intervening between each extinction period.

BAR LOAD IN GRAUS

Fig. 63. Graph showing the relation-
ship of the number of unreinforced reac-
tions performed by albino rats in one hour
to the amount of work involved in each
reaction. (Plotted from data published
by Mowrer and Jones, 16 .)

One group had the bar
weighted throughout the
process exclusively with 5
grams ; the second group had
the bar weighted with 42.5
grams; and the third group
had the bar weighted with
80 grams. The mean number
of unreinforced reactions
made by each of the respec-
tive groups of animals is
shown in Figure 63. There
it may be seen that the num-
ber of extinctive reactions
performed under constant
conditions of habit strength
and motivation is approxi-
mately an inverse linear
function of the work in-
volved in the act. Evidence
reported by Crutchfield (2),
from a rather different ex-

perimental situation, also suggests an inverse linear relationship.
This relationship may be expressed rather precisely by a trans-
position of the expression for I given above, i.e.,

n = I r (B - W)

Since we are committed to a centigrade scale in the present
system, it follows that the maximum of inhibitory potential must
arbitrarily have a value of 100. Accordingly we take the unit of
inhibitory potential as that amount of I R which will just neutralize
one unit of reaction potential. This unit will be called the pav,
a syllable from the name Pavlov. It is suggested that the term pav

INHIBITION— EFFECTIVE REACTION POTENTIAL

28/

be pronounced to rhyme with have. The pav is accordingly defined
in terms of the wat, thus:

1 wat + 1 pav = 0. 1

f Having formally specified the value of I R , we must now inquire
what changes take place in it with the passage of time. It will
be recalled that inhibitory potential is assumed on the submolar
level to have its physical basis in a negative motivational con-
dition or state. This quite probably depends on a substance resi-
dent in the effector organs involved in the response. Now, it is to
be expected that such a substance will gradually be removed by
the blood stream passing through these organs. Moreover, the
amount of this removal per unit time following the cessation of
the action should be proportional to the amount of inhibitory sub-
stance present at any given time. This, of course, is equivalent to
saying that the dissipation of I R will take place according to a
simple decay or negative growth function (p. 199 ff.) of time. We
thus arrive at our third preliminary proposition:

C. Each amount of inhibitory potential ( I R ) diminishes pro-
gressively with the passage of time according to a simple decay
or negative growth function.

THE CONDITIONING OF INHIBITORY POTENTIAL, ITS STIMULUS
GENERALIZATION, AND THE CONCEPT OF EFFECTIVE

REACTION POTENTIAL

At this point in our analysis we need to emphasize a somewhat
different aspect of the above principles from that employed in the
derivation of the preceding two propositions. The new emphasis
will be on that portion of the Mowrer-Miller hypothesis (prelimi-
nary Proposition A) which states that the after-effect of reaction
evocation is a primary negative motivational state or condition.
This means that the after-effects of response evocation in the aggre-
gate constitute a negative drive strongly akin to tissue injury or
“pain.” If this is the case, we should expect that the cessation of
the “nocuous” stimulation in question or the reduction in the

*It must be observed that the formal precision of the definition of the
unit of inhibition is superficially deceptive in that it is programmatic rather
than an accomplished fact. This statement holds for the units of habit
strength and all similar units employed in the present work. It is believed
that the use of such units even in a merely formal and programmatic sense
adds to the clarity of the exposition and will contribute ultimately to an
empirically workable operational definition.

282

PRINCIPLES OF BEHAVIOR

inhibitory substance, or both, would constitute a reinforcing state
of affairs. The response process which would be most closely asso-
ciated with such a reinforcing state of affairs would obviously be
the cessation of the activity itself. In accordance with the “law
of reinforcement” (p. 80 ff.) this cessation of activity would be
conditioned to any afferent stimulus impulses, or stimulus traces,
which chanced to be present at the time the need decrement oc-
curred. Consequently there would arise the somewhat paradoxical
phenomenon of a negative habit, i.e., a habit of not doing some-
thing ( 14 , P- 40 ff.). Thus we arrive at our first corollary:

I. Stimuli closely associated with the acquisition and accumu-
lation of inhibitory potential ( I R ) become conditioned to it in such
a way that when such stimuli later precede or occur simultaneously
with stimulus situations otherwise evoking positive reactions , these
latter excitatory tendencies will be weakened.

Fortunately the existence of such habits is well authenticated,
having long ago been demonstrated experimentally by Pavlov in
what he called conditioned inhibition. Pavlov reports (16, p. 77 ff.)
that a tactile stimulus was conditioned in a dog to a defensive
salivary secretion produced by an injection of weak acid into the
animal’s mouth. In addition, an alimentary conditioned reflex was
set up to the ticking of a metronome by having the latter followed
by feeding. Then the metronome and a neutral stimulus in the
form of a whistle were repeatedly presented together without rein-
forcement, which produced experimental extinction of the condi-
tioned alimentary reaction. Now, according to the present hypoth-
esis this extinction process should become conditioned to any
neutral stimuli associated with it, notably the whistle; the remain-
der of the experiment proved this to be the case. After presenting
the tactile stimulus a couple of times to demonstrate the strength
of its power to evoke the salivary reaction, the experimenter pre-
sented it for the first time in conjunction with the whistle. The
quantitative results are shown in Table 8. There it may be seen
that at 3:16 the tactile stimulus evoked a secretion of 8 drops in
one minute, whereas at 3:25, when the tactile stimulation was
combined with the whistle, the secretion was less than one drop.
At 3:30 the tactile stimulus, now applied alone, evoked 11 drops,
which still further emphasizes the inhibitory effects of the whistle
in the stimulus combination presented at 3:25. It had, of course,
been demonstrated that previous to association with the extinction
of the metronome-salivary conditioned reaction the whistle did not

INHIBITION-EFFECTIVE REACTION POTENTIAL 283

TABLE 8

This Table Presents Experimental Results Illustrating the Action op
Conditioned Inhibition (Evoked by a Whistle) in Largely Suppressing the
Evocation of a Heterogeneous Conditioned Reaction. (From Pavlov, 16 ,
P- 77.)

Time

Stimulus Applied During

1 Minute

Salivary Secretion in Drops
During 1 Minute

3:08 p.m.

tactile

3:16 p.m.

tactile

3:25 p.m.

tactile 4- whistle

less than 1 drop

3:30 p.m.

tactile

produce external inhibition (16, p. 77) of the tactile-evoked sali-
vary reaction.

In continuing the discussion of conditioned inhibition it must
be pointed out that Corollary I has introduced a new dimension
into the concept of inhibitory dynamics; this is to the effect that
the influence of inhibition on behavior evocation may be controlled
by a stimulus. Such a possibility requires the employment of a
new symbol which will explicitly express this fact. We shall do
this simply by adding to the symbol for ordinary inhibitory poten-
tial, I R , the letter S as an extra subscript, thus:

But the moment the action of inhibitory potential ( S I R ) is cued
to a stimulus, the suggestion arises from the analogy to B H R , and
so to g E R , that inhibitory potential will also manifest stimulus gen-
eralization; this brings us to our second corollary:

II. Conditioned inhibitory potential ( a I R ) will manifest stim-
ulus generalization in a manner exactly analogous to that of reac-
tion potentiality ( 8 E R ), as given in Postulate 5.

The appearance of two forms of inhibition on the theoretical
scene instantly raises the question as to how they combine, which,
in turn, requires the introduction of the concept of total inhibitory
potential; this will be represented by the symbol I R . In this way
we arrive at the statement of our fourth preliminary proposition:

D. Simple reactive inhibition (I R ) and conditioned inhibition
(gI R ) summate functionally to produce I R as would corresponding
amounts of habit strength (p. 223).

With the concept of total inhibitory potential available, it be-
comes necessary to introduce explicitly the concept of effective reac-
tion potential; this is represented by the symbol S W R . The intro-

PRINCIPLES OF BEHAVIOR

duction of this concept brings us to the statement of our fifth pre-
liminary proposition:

E. The effective reaction potential (b^r), that reaction
potential which is actually available for the evocation of action
(R), is the reaction potential ( g E R ) less the total inhibitory poten -
tiaX (/*).

Finally, with Proposition E available it becomes possible to
derive the empirically known law of spontaneous recovery. This
concept was employed by Pavlov (16, p. 58) to designate the well-
known fact (see p. 271) that conditioned reactions which had suf-
fered experimental extinction tended in the course of time spon-
taneously to recover a considerable proportion of their original
effective reaction potentiality. By Proposition C, the inhibitory
potential ( I R ) operating against any given response (72) disinte-
grates according to a simple negative growth function. But since
(Proposition D) f B is a summation of I a and B I R , and since (Propo-
sition E),

it follows that B E R will increase as 1 R decreases, which brings us
to our third corollary:

III. Other things equal, an effective reaction potential (b'Er)
which has been reduced by the accumulation of inhibitory potential
(I R ) will recover spontaneously through the mere passage of time,
the course of the recovery being a simple positive growth function
of the time elapsing since the termination of the final response of
the series which produced the inhibition in question.

Ample verification of Corollary III in the case of extinctive
inhibition was seen above (p. 270 ff. and Figure 62).

THE INCOMPLETENESS OF THE STIMULUS GENERALIZATION AND
OF THE SPONTANEOUS RECOVERY OF EXTINCTIVE INHIBITION

Now let it be assumed explicitly, as was done tacitly in the
derivation of Corollary I, that whereas b Ir involves the whole
neural receptor-effector mechanism of habit, I R involves only (or
mainly) the effector portion of this mechanism. Accordingly it is
to be expected that 1 R would not manifest the generalization gradi-
ent characteristic of habit, but would display its presence by a
constant amount of diminished effective reaction potential of the
effector regardless of the stimulus evoking the reaction. This con-

INHIBITION-EFFECTIVE REACTION POTENTIAL 285

stant or non-diminishing amount of reactive inhibition at all points
on the stimulus generalization gradient of inhibitory potential
brings us to our fourth corollary:

IV. When a habit has been set up by well-distributed reinforce-
ments and extinguished by massed evocations, the asymptote or
limit of rise of effective reaction potential ( B E R ), due to the stim-
ulus generalization of the conditioned inhibition, will always be less
than the strength of the effective reaction potential just previous to
the extinction.

A second implication arising from the differential characteristics
of Jr and B I R hinges on the empirically established principle that
in animal experimentation habits are relatively immune to for-
getting, whether set up by the conditioned reaction technique or
by selective learning (7). This means that (Proposition C) I R
will manifest the phenomenon of spontaneous dissipation as a
function of time, whereas B I R , being a habit, will not to any great
extent. These considerations, coupled with Corollary III and the
equation,

Jr = Jr "b sJbj 1

bring us to the conclusion that spontaneous recovery of JE R will
occur only in so far as the extinctive inhibition is comprised of
Jr. Thus we arrive at our fifth corollary:

V. In case a reaction potential ( B E R ) has been set up by dis-
tributed reinforcements and extinguished by massed evocations,
spontaneous recovery of B E R will be incomplete. This we have
seen to be true in the case of Ellson’s investigation (lower curve,
Figure 62).

But since stimulus generalization is based on B I R or the habit
aspect of reactive inhibition, it is not to be expected that the
purely stimulus generalization of extinctive inhibition would show
spontaneous recovery; this brings us to our sixth corollary:

VI. Where effective reaction potential ( B E R ) falls below simple
reaction potential ( B E R ) by reason of purely stimulus generaliza-
tion, i.e., from the action of conditioned inhibitory potential ( B I R ),
the effective reaction potential ( B E R ) will display no spontaneous
recovery whatever . No evidence bearing directly on Corollary VT
has been found. An experimental test of its soundness would

1 The sign + indicates physiological summation according to equations 32
and 43.

286

PRINCIPLES OF BEHAVIOR

accordingly be of special value in determining the validity of the
numerous assumptions ultimately involved in its derivation.

If, however, the difference between B E R and qEr is due to purely
reactive inhibition, i.e., to the action of I R , then complete spon-
taneous recovery should take place; which brings us to our seventh
corollary:

VII. Where effective reaction potential ( B E R ) falls below sim-
ple reaction potential ( B E R ) by reason of the purely response gen-
eralization of extinctive inhibition, spontaneous recovery will occur
and it will be complete .

The spontaneous recovery observed in one portion of Ellson’s
experiment ( 3 ), that represented by the upper curve of Figure 62
(p. 271), presumably took place under conditions which approached
those of response generalization. Accordingly it may, or may not,
be significant for the validity of Corollary VII that spontaneous
recovery was approximately perfect.

THE SUCCESSIVE EXTINCTION OF THE SAME REACTION

POTENTIAL

From Corollary I it follows that I R and B I R are generated con-
currently. In cases where experimental extinction is complete, i.e.,
where (Propositions D and E ),

s E r = 3 E r — ( I R 4- sIr) — 0,

it follows that

sE B ~ Ir 4 " s!r»

This raises many intriguing questions as to the relative amounts
of the two supposed forms of inhibitory potential that are gen-
erated under various conditions. As yet little experimental evi-
dence concerning this matter exists, although Ellson’s results sug-
gest that during the initial extinction the two forms of inhibition
are generated in roughly equal amounts.

From the preceding considerations it may be concluded that if
an organism is subjected to massed extinctive evocations every five
or six hours, say, there will be an appreciable amount of both I R
and B I R generated in the first extinction. Six hours later the Is
will largely have been dissipated, but, since habits do not disin-
tegrate with the mere passage of time, the B I R will remain, 30 there
will be an appreciably diminished amount of reaction potentiality

INHIBITION-EFFECTIVE REACTION POTENTIAL 287

available for a second extinction; this will again generate both I R
and a I R . But since there will be less 8 E R to be extinguished, less
of both I R and S I R will be generated than on the first occasion, so
that fewer extinction evocations and less time will be required.
Since there will be less and less I R generated at each new extinction,
there will be less and less spontaneous recovery after each recovery
period. Thus we arrive at our eighth corollary:

VIII. In case a reaction tendency ( 8 E R ) is subjected to the
same criterion of experimental extinction by massed evocations at
uniform intervals , the amount of spontaneous recovery manifest at
each successive extinction will progressively diminish until ulti-
mately there may be no spontaneous recovery whatever, the number
of unreinforced evocations required to produce a given degree of
experimental extinction on the successive occasions being approxi-
mately a negative growth function of the ordinal number of the
extinction in question.

The evidence bearing on the soundness of Corollary VIII ap-
pears to be internally inconsistent. The necessities of adaptive
dynamics demand that an organism shall not continue forever to
waste its energy performing acts which yield no need reduction;
this is in agreement with the general observation that organisms
do in fact ultimately give up completely the performance of such
acts. On the other hand, there appear to be situations, notably
certain ones involving secondary reinforcement (p. 84 ff.), in
which a considerable amount of spontaneous recovery continues to
occur very many times. A striking case in point has been reported
by Fitts (4). This investigator extinguished rats on a Skinner bar-
pressing habit repeatedly at intervals of a week or longer. He
found that the first three extinctions followed approximately the
course deduced above, but the fourth extinction showed a statis-
tically reliable increase in recovery which persisted, though with
a gradual diminution, for five further extinctions. It seems likely
that this reversal in Fitts’ curve is due to some complex secondary
reinforcement mechanism involving the stimuli arising from frac-
tional antedating goal reactions, though the mechanism itself has
not yet been worked out in detail.

THE PHENOMENON OF DISINHIBITION

If it is true, as implied by Corollary I, that conditioned inhibi-
tion (sic) is a negative habit, it is to be expected that the occur-

288

PRINCIPLES OF BEHAVIOR

rence of an extra stimulus (an unaccustomed stimulus element)
in the stimulus compound (S) would, through afferent neural inter-
action (p. 42 ff.), produce a diminution in the si it- This kind of
causal mechanism is, of course, exactly that which presumably
gives rise to what Pavlov calls external inhibition (see p. 217 ff.).
When applied to inhibition itself, Pavlov calls the action of the
extra stimulus disinhibition (16, p. 61 ff.). This brings us to our
ninth corollary:

IX. Whenever a stimulus element not customarily present in
a compound stimulus ( S ) conditioned to an inhibitory tendency
( gI R ) occurs in such a compound , the amount of inhibitory potential
evokable by the new combination will be less than that normally
evoked by the stimulus compound originally conditioned to the
inhibition.

Since any change in the inhibitory potential can become mani-
fest only indirectly through positive action of some sort, it follows
from the relationship,

sEr — sE r — ( Ir 4" sIr),

that if the extra stimulus produced the same amount of reduction
in b E r as in g I R , the two effects would exactly offset each other;
i.e., no change whatever could occur in bEr, and therefore no
change in observable behavior could result. Nevertheless, disin-
hibition is a well-authenticated empirical phenomenon. This seems
to require the assumption of a special susceptibility of conditioned
inhibition to being upset by extra stimuli. In this connection Pav-
lov remarks that

. . . the inhibitory process is more labile and more easily affected than
the excitatory process, being influenced by st imuli of much weaker physio-
logical strength. (16, p. 99.)

He cites experimental evidence which purports to substantiate this
view; i.e., which shows that a weak stimulus will weaken the bIr
only, whereas a stimulation which includes a strong “extra” stim-
ulus element may evoke no reaction whatever since it would also
abolish the bHr, and so the g E R , by “external” inhibition.

If Pavlov’s assumption of the differential action of weak extra
stimuli on bIr and bEr is accepted, an important implication fol-
lows at once from the relationship,

sEr — sEr — (I R sIr)-

Thus we arrive at our tenth corollary:

INHIBITION— EFFECTIVE REACTION POTENTIAL 289

X. If a reaction 'potential ( 8 E R ) has been partially or wholly
extinguished, the inclusion of a mild extra stimulus in the condi-
tioned stimulus compound (S) will result in the strengthening of
the effective reaction potentiality ( 8 E R ).

But since 8 Ir constitutes only a portion of the total inhibition
( Ir ) which weakens 8 E R , it follows that disinhibition can only par-
tially restore 8 E R to the original value of 8 E R ; this leads to our
eleventh corollary:

XI. When an excitatory habit has been set up by means of
well-distributed reinforcements and has then been extinguished,
the most effective disinhibitory stimulus possible will never ( except
through “oscillation” — see p. 304 ff.) enable S to evoke an R with
as great vigor, certainty, or speed as before the extinction occurred.

INHIBITION OF REINFORCEMENT

It is evident from the above version of the Mowrer-Miller
hypothesis (Proposition A) that reactive inhibition must be gen-
erated whenever reactions are evoked, whether reinforcement occurs
or not. If reinforcement does not occur, the inhibitory potential
generated by the response may be called extinctive inhibition; the
inhibition generated if the response is followed by reinforcement
has been called by Hovland, inhibition of reinforcement ( 8 ), from
the circumstances of its origin, even though the inhibition is pre-
sumably not dependent upon the reinforcement process. If rein-
forcement occurs, the consequent increase in habit strength ( 8 H R )
will so increase the reaction potential ( 8 E R ) that when the in-
hibitory potential ( I R ) arising from the successive reactions is
deducted, the effective reaction potential ( 8 E R ) will usually be
superthreshold in amount, i.e., more than enough, given normal
stimulation and motivation, to evoke the reaction.

Now, inhibitory potential can be observed only indirectly
through the failure to occur of some positive reactions which the
antecedent conditions would otherwise produce. Because of this
fact and of the usual over-riding effects of reinforcement, it hap-
pens that inhibition of reinforcement usually does not manifest
itself in any very dramatic manner such, for example, as in the
total cessation of reaction evocation so characteristic of experi-
mental extinction. Probably it is because of these circumstances
that relatively few investigators have noticed it. As might be ex-

PRINCIPLES OF BEHAVIOR

290

pected, it was Pavlov who seems first to have observed and de-
scribed this phenomenon. He remarks:

The development of inhibition in the case of conditioned reflexes which
remain without reinforcement must be considered only as a special in-
stance of a more general case, since a state of inhibition can develop also
when the conditioned reflexes are reinforced. The cortical cells under the
influence of the conditioned stimulus always tend to pass, though some-
times very slowly, into a state of inhibition. . . . This inhibition is of the
same character as the internal inhibition which has been described in
previous lectures, and it exhibits the same properties of irradiating to
other cortical elements which are not primarily involved. {16, pp. 234-
248)

In the above context Pavlov describes an illustrative experiment
in which a conditioned reflex in the course of a number of rein-

TRIALS

Fia. 64. Graphs showing the course of the acquisition of the conditioned
lid reflex as a function of the time interval separating the reinforcements.

(From Calvin, 1 , as presented by Hilgard and Marquis, 7 , p. 149.)

forcements given rather close together actually diminished to a
zero strength (see Figure 64).

Thus we arrive at our twelfth and thirteenth corollaries:

XII. Whenever conditioned reactions are evoked, whether rein-
forced or not, reactive inhibition (I R ) is generated.

XIII. In the case of closely massed reinforcements, the curve
of acquisition of effective excitatory potential (s^r), particularly

INHIBITION-EFFECTIVE REACTION POTENTIAL

291

in its later stages , will be distorted by inhibition of reinforcement
below the learning curve of bEr } in extreme cases showing an actual
fall with continued practice.

Calvin reports such a curve (Figure 64), in which the rein-
forcements occurred at the rate of eighteen times per minute; this
curve not only ceases to rise after about 25 reinforcements, but
actually shows a slight fall.

DISINHIBITION AND THE INITIAL, RISE IN THE CURVE OF

EXPERIMENTAL EXTINCTION

In 1930 Switzer (18) reported a novel form of experimental ex-
tinction curve, based on the extinction of the conditioned lid reac-
tion. Instead of falling
abruptly from the initial un-
reinforced reaction as in Fig-
ure 57, the curve showed at
first a sharp rise in ampli-
tude of reaction (Figure 65).

After one or two more unre-
inforced reactions this initial
rise was followed by the fall
usually encountered in the
extinction process. Hudgins
(9; 10, p. 439) and others
have fully verified Switzer’s
original discovery. Hovland
(5) has reported the out-
come of an ingenious experi-
ment which he interprets as
explaining the phenomenon
found by Switzer. The habit
involved was a galvanic skin
reaction produced by an electric shock which had been conditioned
to an auditory stimulus. The results of the experiment are shown
concisely by the four graphs appearing in Figure 66, graphs B and
C being of special significance. Graph B, which is a curve of ex-
perimental extinction following massed reinforcements, shows what
purports to be the Switzer effect; graph C is a curve of experimental
extinction following a form of distributed reinforcements and shows
a close approximation to the conventional curve of extinction. Hov-

Fig. 65. Graph showing the Switzer
phenomenon. Note the initial rise in the
composite curve of experimental extinc-
tion of a conditioned eyelid reaction.
(Plotted from pooled values published by
Switzer, 18, p. 86.)

2 9 2

PRINCIPLES OF BEHAVIOR

land interprets these results as indicating (a) that massed rein-
forcements in considerable numbers leave at their conclusion a rela-
tively large amount of “inhibition of reinforcement”; (b) that the
transition from reinforcement to non-reinforcement acts as a disin-
hibiting agent, producing disinhibition of the accumulated inhibi-

extinction trial extinction trial

Fio. 66. Extinction curves following various conditions of reinforcement.
Results plotted in terms of ratios (in per cent) of responses on successive
extinction trials to response on first extinction trial. (A) 8 reinforcements.
Extinction immediately. (£) 24 reinforcements. Extinction immediately.
(C) 24 reinforcements, distributed into 3 groups of 8 each. Rest period of
30 minutes between groups. Extinction immediately after last group of
reinforcements. ( D ) 24 reinforcements. Extinction 30 minutes after last
reinforcement. (Reproduced from Ho viand, 8, p. 431.)

tion of reinforcement. This accounts for the initial rise shown in
graph B.

The set of assumptions outlined in the preceding pages of
the present chapter imply the initial rise of the curve of experi-
mental extinction in the following manner: (a) massed reinforce-
ments generate a relatively large amount of I R and consequently
a strong negative drive; (b) each pause between reinforcements,
even if brief, produces a slight reduction in the need (for inactivity,
or rest) ; (c) this serves as a reinforcing state of affairs setting up

INHIBITION-EFFECTIVE REACTION POTENTIAL,

293

a certain amount of conditioned inhibition { b Ir) ; (d) the sudden
transition from reinforcement to non-reinforcement, on the first
non-reinforced stimulation, withdraws from the afferent impulses
which customarily were present at previous reinforcements, the
stimulus traces of the shock; (e) this change in the make-up of the
afferent compound, through the principle of afferent interaction (p.
42 ff.), is sufficient to produce disinhibition of the conditioned
inhibition of reinforcement; which (/) results in the initial rise in
the curve of experimental extinction. Thus we arrive at our four-
teenth corollary:

XIV. When conditioned reactions are set up by means of
massed reinforcements, conditioned inhibition is generated which,
at the outset of extinction, is disinhibited through the change in the
functioning afferent impulses, with the result that the curve of
experimental extinction shows an initial rise.

While Corollary XIV agrees in the main with Hovland’s experi-
mental findings, there are certain respects in which it does not.
The greatest single inconsistency is shown in graph D, which rep-
resents the extinction of a conditioned reaction based on 24 massed
reinforcements, 30 minutes after the conclusion of the reinforcement
series. Since disinhibition is assumed to be based on B I B} which
is regarded as a habit, and since ordinary habits do not disinte-
grate appreciably in 30 minutes (7), there should have been about
as much B I R to be disinhibited in the case of graph D as in that
of graph B, which clearly is not the case. The discrepancy is of
considerable importance, since it points to a serious defect in the
postulates which generate Corollary XIV. This matter clearly
needs further intensive experimental investigation, particularly
from the point of view of the assumed B I R .

THE LAW OP LESS WORlK

One of the traditional methods employed in the investigation
of the gradient of reinforcement (p. 137 ff.) has been to give an
organism the choice of two paths to the attainment of some sort
of reinforcing agent, such as food, and study its behavior as rein-
forcements of the two behavior sequences accumulate. In one of
the best and most recent investigations of this kind, Grice (£) con-
cludes in effect that if the temporal factor (upon which the gradi-
ent of reinforcement is formulated) were completely equalized,
there would still be a marked preference for the shorter path. In

PRINCIPLES OF BEHAVIOR

294

the light of Corollary I and the concept of inhibition of reinforce-
ment, it may be shown that the outcome suggested by Grice would
be inevitable. Let it be supposed that, as in one part of Grice’s
experiment, a rat is given a choice of two paths to food, one 6 feet
in length and one 12 feet in length. It is clear that much more
work must be performed in traversing 12 feet than in traversing
6 feet. By Propositions A and B it follows that the amount of
inhibitory potential generated in traversing the 12-foot path will
be greater than that generated in traversing the 6-foot path. As-
suming the temporal delay in reinforcement to be constant for both
paths, and therefore the two habit strengths ( 8 H R ) to be constant,
it follows (p. 253 ff.) that the reaction potential ( 8 Er) must be
constant under the same degree of drive, e.g., hunger. Now accord-
ing to Proposition F y the effective reaction potential, that available
for the actual evocation of reaction, 8 Er, is the difference between
the actual reaction potential ( 8 E R ) and the total inhibitory poten-
tial ( Ir ). But since the I R generated by traversing the long path
is greater than that generated by traversing the short path, it fol-
lows by Corollary I that the conditioned inhibition ( a I R ) against
the long path will be greater than that against the short path. As
a consequence the effective reaction potential for the short path
must be greater. The organism will accordingly come to choose
this path to the point of reinforcement.

We thus arrive at our fifteenth corollary:

XV. If two or more behavior sequences , each involving a dif-
ferent amount of energy consumption or work (W), have been
equally well reinforced an equal number of times , the organism
will gradually learn to choose the less laborious behavior sequence
leading to the attainment of the reinforcing state of affairs.

This corollary applies to a very extensive range of phenomena
subsumable under the law of less work , which within recent years
has attracted the attention of a number of writers. Gengerelli (5),
Wheeler (23) , Tsai (20), Waters (22), and Crutchfield (2), among
others, have referred to this in various terms such as the “law of
minimal effort,” and the “law of least action.” There has been
a tendency on the part of some to regard the law of less work as
a primary or basic principle, perhaps as a special case of the law
of least action employed in the theory of physical dynamics. In
the present system, Corollary XV, which in effect is a statement
of the law of less work, has the status of a secondary or derived
principle. Moreover, it is definitely a behavioral law conforming

INHIBITION-EFFECTIVE REACTION POTENTIAL 295

to the principles of learning. As such the lower limit of the work
involved in the attainment of any given reinforcement is not abso-
lute but is dependent on the range of behavior of which the habit
system and the effector equipment of the organism are capable.
The law of less work is accordingly a relative and derived principle
rather than an absolute and primary one, as in physical dynamics.

As a further indication of the non-mechanical nature of the
law of less work we may state without elaboration two implica-
tions flowing from it in conjunction with certain other principles
of the present system:

XVI. The rate of acquisition of the preference for the behavior
sequences involving the less energy expenditure or work will be
faster with massed than with distributed reinforcements.

XVII. The preference acquired by massed reinforcements for
the behavior sequence involving less work will lose some , but not
all y of its strength as the result of the passage of a few hours of
no practice. The advantage not lost through the passage of time
would be due to that portion of I R which is constituted by s Ir;
this, being a habit, will not spontaneously dissipate very rapidly.

THE PHENOMENON OF REMINISCENCE AND THE ECONOMY OF

DISTRIBUTED REINFORCEMENTS

Let it be supposed that a considerable number of reinforcements
associating a stimulus and a reaction have occurred in close succes-
sion and that an interval of some hours elapses before a test is
made of the reaction evocation power of the stimulus. By the
law of reinforcement, a considerable habit strength will have been
produced through the reinforcements, and (Corollary XII) a con-
siderable amount of inhibition of reinforcement ( I R ) will also have
developed through the incidental reaction evocations. This in-
hibitory potential (Corollary XIII) will have reduced correspond-
ingly the effective potentiality of reaction evocation (bEr). But
(Proposition C) the lapse of some hours will permit considerable
spontaneous recovery from the inhibitory potential, which (Corol-
lary III) will increase the strength of effective reaction potentiality
( b & r ). Now, conditioned reactions and most habits set up experi-
mentally in animals (7) show very little loss (“forgetting”) through
the passage of time; and in the case of rote learning in humans,
the loss of g H R from forgetting appears to be less rapid (11, p. 121)
than that of I R . It follows that in all these cases, after the period

PRINCIPLES OF BEHAVIOR

296

of no practice there will exist a net increased effective reaction
potentiality ( a E R ) ; this increase has received the not too appro-
priate designation of reminiscence. Thus we arrive at our eight-
eenth, nineteenth, twentieth, and twenty-first corollaries:

XVIII. In case a simple conditioned reaction is set up to an
appreciable strength by massed practice and the final reinforce-
ment is followed by a no-practice period several times as long as
the interval between reinforcements , after which the stimulus is
again delivered , motivation remaining constant , the reaction- evoca-
tion potentiality of this stimulus uill be greater than it was at the
termination of the original reinforcement sequence.

XIX. In the case of simple conditioned reactions, reminiscence
if plotted as a function of time will approximate a simple growth
function.

XX. In the case of rote series learned by massed practice,
reminiscence will rise at first with a negative acceleration, which
will presently be replaced by a fall (11, pp. 261, 263, 277).

XXI. The closer the massing of the reinforcements in the origi-
nal learning, the greater will be the extent of the reminiscence
effect.

The phenomenon of reminiscence in the conditioned reflex set
up by means of massed practice was, as usual, first described by
Pavlov (16, p. 249). A rather elaborate quantitative investigation
of the phenomenon with human subjects, incidental to the study
summarized in Figure 64, has been reported by Calvin (1). He
found after a no-practice period of 24 hours a mean increase of
from 5.75 reactions per ten trials to 7.10, a gain of 1.35 points,
where the reinforcements were three per minute; where the rein-
forcements were eighteen per minute he found an increase of from
2.30 to 7.25 reactions, a gain of 4.95 points. This shows an appre-
ciable advantage in reminiscence for the more closely massed rein-
forcements.

The phenomenon of reminiscence in rote learning has been
known for many years. The last and most precise investigation
of reminiscence in this form of learning was reported by Ward (21).

If we apply Corollary XVIII to any learning situation, it is
evident that when reinforcements are separated by time intervals
of moderate length, the greater these intervals, the greater will be
the amount of spontaneous recovery during the intervals, i.e., the
less the aggregate inhibitory potential at the end of learning, and
therefore the greater will be the effective habit strength at the

INHIBITION-EFFECTIVE REACTION POTENTIAL

2 97

conclusion of the last reinforcement. This is known in learning
literature as “the economy of distributed repetitions.” An excellent
demonstration of this principle is furnished by the Calvin study (1 )
just referred to. Using human subjects, this investigator condi-
tioned to a light stimulus the lid-closure reflex originally evoked by
a shock below the eye. One group of 20 subjects received rein-
forcements at the rate of three per minute, another group at the
rate of nine per minute, and a third group at the rate of eighteen
per minute. The course of the learning in terms of the per cent
of presentations of the conditioned stimulus evoking the conditioned
reaction was as indicated in Figure 64. These curves show that
the rate of learning where three reinforcements were given per
minute was much faster than where reinforcements were given more
closely massed. Many parallel experiments in various fields of
learning, particularly in the rote learning of nonsense syllables
(11, p. 127 ff.), have demonstrated the same type of economy.

From the preceding considerations we accordingly formulate our
twenty-second corollary:

XXII. Within limits, the greater the time interval separating
the reinforcements of learning, the greater will be the effective ex-
citatory potential (bEr) at the conclusion of the last reinforcement.

SUMMARY

The Mowrer-Miller hypothesis states in effect that all responses
leave behind in the physical structures involved in the evocation,
a state or substance which acts directly to inhibit the evocation of
the activity in question. The hypothetical inhibitory condition or
substance is observable only through its effect upon positive reac-
tion potentials. This negative action is here called reactive inhibi-
tion. An increment of reactive inhibition (A I R ) is assumed to be
generated by every repetition of the response (72), whether rein-
forced or not, and these increments are assumed to accumulate
except as they spontaneously disintegrate with the passage of time.
The magnitude of the individual increments, and therefore of the
rate of accumulation, appears clearly to be in part a positively
accelerated increasing function of the amount of energy consumed
by the response.

Because of the motivational characteristics of reactive inhibi-
tion, or inhibitory potential, it is opposed to reaction potential
(bEr) rather than to habit ( g H R ), as is sometimes supposed. Thus

298 PRINCIPLES OF BEHAVIOR

effective reaction potential ( B E R ) y the potential actually available
for the evocation of action, is the reaction potential less the in-
hibitory potential.

Since under ordinary learning conditions response and rein-
forcement occur in parallel, the strengthening of the habit due to
reinforcement usually is great enough to over-ride the accumulating
inhibition. As a consequence, inhibition of reinforcement is only
detected by special means. In case little or no reinforcement fol-
lows the reaction evocations, extinctive inhibition soon neutralizes
the reaction potential, the stimulus gradually ceases to evoke the
response, and there ensues the state known as experimental extinc-
tion which thus appears as a secondary or derived phenomenon.

The Mowrer-Miller hypothesis regards reactive inhibition as
essentially a need to cease action, i.e., a need for rest; it follows
that anything which reduces this need should serve as a reinforcing
state of affairs. Since the cessation of action reduces the afferent
proprioceptive impulses generated by it in the presence of the
inhibiting condition, particularly when many responses have gen-
erated a considerable amount of inhibition, it comes about that the
cessation of action, rather than action, becomes conditioned to
whatever stimuli may be present. In this way we find a plausible
explanation of conditioned inhibition ( B I R ) and of the stimulus
generalization of extinction effects. There are a number of indi-
cations that phenomena analogous to conditioned inhibition and
stimulus generalization of inhibition occur under conditions of
ordinary learning reinforcements, though not all the empirical evi-
dence harmonizes with this a priori expectation. For this reason
the theory of the origin of a I R must be regarded with somewhat
more than the usual amount of distrust.

Because conditioned inhibition ( B I R ) is generated as a secondary
effect from the accumulation of reactive inhibition (I R ) f it follows
that at least in extinction situations both I R and B I R will result.
Assuming that the two summate physiologically, it follows that at
complete experimental extinction the excitatory potential { b Er)
will be opposed or neutralized in part by 1 R and in part by gin.
Now, I R dissipates spontaneously through the passage of time, but
sI Rf being a true habit, presumably does not, at least to any great
extent. The dissipation of I R will produce spontaneous recovery of
direct extinction effects, but this will naturally result in only partial
recovery. On the other hand the second inhibitory component m

INHIBITION— EFFECTIVE REACTION POTENTIAL

299

extinction ( 8 Ib) should be subject to external inhibition. Since
bIb is responsible for only a portion of the depression of 8 Er below
bE r , disinhibition, which presumably operates only on 8 I R , also
should never produce complete recovery. The slight initial rise in
response vigor when extinction follows massed reinforcements is
plausibly interpreted as the external inhibition of the conditioned
inhibition ( b Ir ) presumably set up during the reinforcement proc-
ess. The facts, however, are not wholly in harmony with this
interpretation.

In case a reaction tendency ( 8 E R ) is extinguished a good many
times, each extinction being performed by massed practice on sepa-
rate occasions, the gradual accumulation of the relatively perma-
nent conditioned inhibition implies that the time required for the
successive extinctions of the reaction tendency should grow less and
less, the minimum approaching zero as a limit.

The magnitude of Ir, and also, presumably, of 8 Ir> generated
by a given number of response evocations depends upon the amount
of energy consumption or work (W) involved. This implies that
of two or more alternative behavior sequences repeatedly executed
by the organism in the attainment of an ordinary reinforcement,
that sequence will finally come to be chosen which involves the
less work or the less tissue injury. This is the important “law of
less work,” which, as pointed out by James {12), accounts for the
prevalence of “laziness” in the behavior of organisms.

Because reactive inhibition ( I R ) dissipates spontaneously
through the passage of time, it follows that a part of the “inhibi-
tion of reinforcement” will dissipate during the pauses which occur
throughout learning by distributed reinforcements or repetitions.
The less the l Ry presumably, the less will be the sI Rf and so, cer-
tainly, the less the I R and consequently the greater will be the
bE r at the end of the learning process. Thus is explained the
well-known empirical law of the economy of distributed repetitions
in learning. If, on the other hand, a considerable number of rein-
forcements are massed and then a pause occurs, the same principle
leads to the frequently observed empirical phenomenon of spon-
taneous recovery of effective reaction potential (sE R ) known as
“reminiscence.”

In view of the considerations, molar and submolar, put forward
in the preceding pages, we now formulate our eighth and ninth
primary molar principles:

PRINCIPLES OF BEHAVIOR

298

effective reaction potential ( B E R ), the potential actually available
for the evocation of action, is the reaction potential less the in-
hibitory potential.

The Mowrer-Miller hypothesis regards reactive inhibition as
essentially a need to cease action, i.e., a need for rest; it follows
that anything which reduces this need should serve as a reinforcing
state of affairs. Since the cessation of action reduces the afferent
proprioceptive impulses generated by it in the presence of the
inhibiting condition, particularly when many responses have gen-
erated a considerable amount of inhibition, it comes about that the
cessation of action, rather than action, becomes conditioned to
whatever stimuli may be present. In this way we find a plausible
explanation of conditioned inhibition ( 8 I R ) and of the stimulus
generalization of extinction effects. There are a number of indi-
cations that phenomena analogous to conditioned inhibition and
stimulus generalization of inhibition occur under conditions of
ordinary learning reinforcements, though not all the empirical evi-
dence harmonizes with this a priori expectation. For this reason
the theory of the origin of 8 I R must be regarded with somewhat
more than the usual amount of distrust.

Because conditioned inhibition ( a I R ) is generated as a secondary
effect from the accumulation of reactive inhibition (Jr), it follows
that at least in extinction situations both I R and 8 I R wifi result.
Assuming that the two summate physiologically, it follows that at
complete experimental extinction the excitatory potential ( 8 E R )
will be opposed or neutralized in part by I R and in part by 8 Ir>
Now, I R dissipates spontaneously through the passage of time, but
bI*, being a true habit, presumably does not, at least to any great
extent. The dissipation of I R will produce spontaneous recovery of
direct extinction effects, but this will naturally result in only partial
recovery. On the other hand the second inhibitory component in

INHffimON— EFFECTIVE REACTION POTENTIAL 299

extinction { 8 Ir) should be subject to external inhibition. Since
fl/j* is responsible for only a portion of the depression of 8 E R below
bEr, disinhibition, which presumably operates only on 8 I R , also
should never produce complete recovery. The slight initial rise in
response vigor when extinction follows massed reinforcements is
plausibly interpreted as the external inhibition of the conditioned
inhibition ( 8 Ir) presumably set up during the reinforcement proc-
ess. The facts, however, are not wholly in harmony with this
interpretation.

In case a reaction tendency ( g E R ) is extinguished a good many
times, each extinction being performed by massed practice on sepa-
rate occasions, the gradual accumulation of the relatively perma-
nent conditioned inhibition implies that the time required for the
successive extinctions of the reaction tendency should grow less and
less, the minimum approaching zero as a limit.

The magnitude of I R , and also, presumably, of 8 Ir, generated
by a given number of response evocations depends upon the amount
of energy consumption or work (W) involved. This implies that
of two or more alternative behavior sequences repeatedly executed
by the organism in the attainment of an ordinary reinforcement,
that sequence will finally come to be chosen which involves the
less work or the less tissue injury. This is the important “law of
less work,” which, as pointed out by James {12), accounts for the
prevalence of “laziness” in the behavior of organisms.

Because reactive inhibition (/*) dissipates spontaneously
through the passage of time, it follows that a part of the “inhibi-
tion of reinforcement” will dissipate during the pauses which occur
throughout learning by distributed reinforcements or repetitions.
The less the I R , presumably, the less will be the 8 I R , and so, cer-
tainly, the less the I R and consequently the greater will be the
8 E r at the end of the learning process. Thus is explained the
well-known empirical law of the economy of distributed repetitions
in learning. If, on the other hand, a considerable number of rein-
forcements are massed and then a pause occurs, the same principle
leads to the frequently observed empirical phenomenon of spon-
taneous recovery of effective reaction potential ( 8 E R ) known as
“reminiscence.”

In view of the considerations, molar and submolar, put forward
in the preceding pages, we now formulate our eighth and ninth
primary molar principles:

3 00

PRINCIPLES OF BEHAVIOR

POSTULATE 8

Whenever a reaction ( R ) is evoked in an organism there is created as
a result a primary negative drive (Z>) ; (a) this has an innate capacity
(Ir) to inhibit the reaction potentiality (sEr) to that response; (b) the
amount of net inhibition (/«) generated by a sequence of reaction evoca-
tions is a simple linear increasing function of the number of evocations
(n) ; and (c) it is a positively accelerated increasing function of the work
(W) involved in the execution of the response; ( d ) reactive inhibition
(Ir) spontaneously dissipates as a simple negative growth function of
time (*'")•

POSTULATE 9

Stimuli (S) closely associated with the cessation of a response ( R )
(a) become conditioned to the inhibition (Ir) associated with the evoca-
tion of that response, thereby generating conditioned inhibition; (6) con-
ditioned inhibitions (sIr) summate physiologically with reactive inhibi-
tion (Ir) against the reaction potentiality to a given response as positive
habit tendencies summate with each other.

NOTES

(o) rE b

(b)

Mathematical Statement of Postulate 8
1 bE b — Is

n _ s!r(B — W )

rvrhere W =*

(d) ‘" 7 * =

B - W
F’L

(36)

(37)

(38)

(39)

(40)

Mathematical Statement of Postulate 9

(<0 /« = /* + si* - In (41)

(fc) sis = bE s - f B . (42)

The Mowrer-Jones Graph
The equation of the line fitted to these data is,

n — 373 3.2 ffm t

where n is the number of unreinforced reactions required to produce extinction,
and gm is the pressure in grams required at each reaction.

INHIBITION— EFFECTIVE REACTION POTENTIAL 301

It may be noted in this connection that the form of the above equation is
not exactly that shown as (37) above. A transformation of the former equation
to agree with (37) is.

n =

116.6 - W

.3125

where 116.6 = B, and .3125

(43)

Equations Expressing the Laws of the Disintegration of Reactive
Inhibition (I R ) and of the “Recovery" of Effective

Reaction Potential ( s Er)

The negative growth or decay principle according to which reactive inhibition
is assumed to disintegrate (Proposition C) is :

‘"7* = /* X 10-*‘"\

Since,

bE r = bBr — (/« + bIr)»

it follows that &Er after the lapse of time, t'", will be,

‘ sEr — sEr — (*/« - f - bIr)

= bEr - ( Ib • 4 - bIr),

from winch it follows that sEr will with sufficient time “recover" substantially
the amount to which it is depressed by Ir, but not the amount to which it is
depressed by gJ s ; and the recovery will take place according to the exponential
or positive growth function of time (*"')•

Problems Connected with the Conditioning of Inhibitory Potential ( g I R )

The theory of the generation of bIr from Ir presents a number of problems
which probably cannot be cleared up without a considerable amount of co-
ordinated research. One major problem may be stated as follows: If the ces-
sation of contraction can serve as a reinforcing state of affairs, why does this not
serve to set up habits involving muscular contraction, as well as inhibition? Such
an implication seems to be perfectly legitimate, and it presents numerous in-
triguing possibilities. In this connection it is important to recall the nature of the
gradient of reinforcement (p. 139 ff.), which is to the effect that the process most
closely preceding the reinforcing state of affairs will be the one most strongly
reinforced. On this principle, the cessation of the “nocuous" stimulation from a
muscle will reinforce most strongly the cessation, relaxation, or inhibition of the
act which produced the discharges and distinctly less the active contraction which
necessarily has preceded the relaxation. Thus while some reinforcement of
excitation leading to positive reaction potential ( sEr ) would result from the
cessation of “fatigue" stimulations arising from muscular action, a much greater,
and therefore a dominating, amount of conditioned inhibitory potential (bIr)
would be generated.

A second and still more complex problem may be stated as follows: If the
cessation, relaxation, or inhibition of action is susceptible to being conditioned to
stimuli, as the phenomenon of conditioned inhibition in conditioned reflexes

30-

PRINCIPLES OF BEHAVIOR

strongly suggests, why does not an ordinary reinforcing agent, such as the reduc-
tion of the food need, reinforce inhibitory, quite as much as excitatory, tendencies?

While this problem requires a much more thorough examination than can
be given here, a few suggestions may be made. In the first place, there is much
evidence that conditioned inhibition is generated in extinction situations, and
the initial rise of the extinction curve suggests that conditioned inhibition is also
evolved in ordinary reinforcing situations. It is conceivable that some of this
conditioned inhibition is set up by means of the mechanism just described. The
critical question concerns the relative amounts of positive reinforcement ( bHr )
and of negative reinforcement (sIr) which will be generated according to the
present set of hypotheses. If inhibitory tendencies were reinforced the same as
excitatory tendencies, the two might simply neutralize each other, in which case
no positive effective reaction potentials would develop, and effective learning
could not occur. Such an outcome would be implied by the naive supposition
that reinforcement does not take place until after the cessation of the act rein-
forced. For example, in the Skinner procedure the animal does not eat until
after the cessation of the muscular contraction which depresses the bar. Actu-
ally, while this is true of primary reinforcement, it is also true that a large portion
of the reinforcement in such learning is secondary in nature, and this secondary
reinforcement, e.g., the click of the magazine, occurs during the contraction and
; preceding the relaxation. In this connection it is well to recall evidence of the
occurrence of reinforcement when the reinforcing state of affairs precedes the
reaction reinforced. Both the work of Thorndike (19, p. 35) and that of Jenkins
(13, pp. 58a and 72a), however, have shown that the forward wing of this double
gradient is relatively much lower than the backward one. The difference in the
strength of reinforcement in the two positions should give the conditioned excita-
tory tendency (sEr) something like the advantage over the conditioned inhibitory
tendency (sIr) that experiment shows it to have in fact.

This question will evidently require much further investigation before a
confident decision can safely be made regarding numerous aspects of reactive
inhibition in adaptive behavior. Indeed, the present chapter may be considered
to be largely an analytical exploration of a very rich field, preliminary to such
a coordinated research program. It is also to be remembered that this uncer-
tainty essentially concerns the submolar basis of Postulate 9; from a logical
point of view all of this difficulty is eliminated when we take this proposition as a
primary molar law, i.e., as a separate postulate. If, as seems likely, Postulate 9
is later rigorously derivable from other principles of the system, it will become a
corollary, and the number of primary principles will thereby be reduced by one.

REFERENCES

1. Calvin, J. S. Decremental factors in conditioned-response learning. PhX).

thesis, Yale University, 1939.

2. Crutchfield, R. S. The determiners of energy expenditure in string-

pulling by the rat. J. Psychol., 1939, 7, 163-178.

3. Ellson, D. G. Quantitative studies of the interaction of simple habits.

I. Recovery from specific and generalized effects of extinction. J. Exper.
Psychol, 1938, S3, 339-358.

4. Fitt8, P. M. Perseveration of non-rewarded behavior in relation to food-

deprivation and work-requirement. J. Genet. Psychol ^ 1940, 57, 165-191-

INHIBITION-EFFECTIVE REACTION POTENTIAL 303

5. Gengerelli, J. A. The principle of maxima and minima in animal learn-

ing. J. Comp. Psychol., 1930, 11, 193-236.

6 . Grice, G. R. An experimental study of the gradient of reinforcement in

maze learning. J. Exper. Psychol., 1942, 80, 475-489.

7. Hilgard, E. R., and Marquis, D. G. Conditioning and learning. New

York: D. Appleton-Century Co., Inc., 1940.

8. Hovland, C. I. ‘Inhibition of reinforcement’ and phenomena of experi-

mental extinction. Proc. Natl. Acad. Sci., 1936, 22, No. 6 , 430-433.

9. Hudgins, C. V. Conditioning and the voluntary control of the pupillary

light reflex. J. Gen. Psychol., 1933, 8, 3-51.

10. Hull, C. L. Learning: II. The factor of the conditioned reflex. Chapter

9 in A handbook of general experimental psychology . Worcester, Mass.:
Clark Univ. Press, 1934.

11. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. Mathematico-deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

12. James, W. T. Experimental observations indicating the significance of

work on conditioned motor reactions. J. Comp. Psychol., 1941, 82, 353-
366.

13. Jenkins, W. O. Studies in the spread of effect. PhJD. thesis, Yale Uni-

versity, 1942.

14. Miller, N. E., and Dollard, J. Social learning and imitation. New

Haven: Yale Univ. Press, 1941.

15. Mowrer, O. H., and Jones, H. M. Extinction and behavior variability as

functions of effortfulness of task (in press).

16. Pavi/iv, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

17. Robinson, E. S. Work of the integrated organism. Chapter 12 in A

handbook of general experimental psychology. Worcester, Mass.: Clark
Univ. Press, 1934.

18. Switzer, S. A. Backward conditioning of the lid reflex. J. Exper.

Psychol., 1930, IS, 76-97.

19. Thorndike, E. L. The psychology of wants, interests, and attitudes.

New York: D. Appleton-Century Co., Inc., 1935.

20. Tsai, L. S. The laws of minimum effort and maximum satisfaction in

animal behavior. Monog. Natl. Res. Instit. Psychol. (Peiping, China),
1932, No. 1. (See in Psychol. Abslr., 1932 6, 4329.)

21. Ward, L. B. Reminiscence and rote learning. Psychol. Monog., 1937, 49,

No. 4.

22. Waters, R. H. The principle of least effort in learning. J. Gen. Psychol.,

1937, 16, 3-20.

23. Wheeler, R. H. The science of psychology. New York: Thomas Y.

Crowell Co., 1929.

CHAPTER XVII

Behavioral Oscillation

It is an everyday observation that organisms vary in their
performance even of well-established, habitual acts from occasion
to occasion and from instant to instant on the same occasion. We
are able to recall a name at one time but not at another; in shoot-
ing at a target we ring the bell at one shot, but not at the next;
and so on.

When a six- place number is divided by a five-place number
repeatedly by an automatic calculating machine, the exact identity
of the quotient obtained on all the occasions is taken for granted;
if the average person were to perform the same divisions, using
pencil and paper, he would consider himself lucky if exactly the
same result were obtained each time. While first-rate calculating
machines sometimes get out of order and make errors, ordinary
inorganic mechanisms under the same external conditions show, m
general, much less variability in behavior than do organisms. In-
deed, variability, inconsistency, and specific unpredictability of
behavior have long been recognized as the chief molar distinctions
between organisms and inorganic machines. Clearly a character-
istic so fundamental as this must find an important place in any
adequate theory of organismic behavior.

EXPERIMENTAL DEMONSTRATIONS OF BEHAVIORAL OSCILLATION

Even when the strength of a reaction potential has become
stabilized at a value well above the reaction threshold, and the
conditioned stimulus evokes its reaction with a considerable degree
of consistency, both the amplitude and the latency of the reaction
always oscillate from trial to trial. This is illustrated nicely by
data from an unpublished study performed in the author’s labora-
tory by Ruth Hays and Charles B. Woodbury. In this investi-
gation a hungry albino rat was placed in a Skinner type of appa-
ratus (Figure 61, p. 268), the single pressure bar of which was
provided with a recording dynamometer. In the first part of the
experiment the apparatus was so set that the animal was required
to make a pressure of 21 grams before the food-pellet reward would

304

BEHAVIORAL OSCILLATION

3<>5

be delivered. Four days of training were then given, on each of
which the animal received 100 reinforcements; on the day following,
under exactly the same conditions, the animal made the distribu-

13* 17* 21- 25 - 29 - 33- 37- 41- 45 - 49 - 53 - 57-

INTENSITY OF PRESSURES IN GRAMS

Flo. 67. Two reaction-intensity distributions of a rat in a bar-pressing ex-
periment, designed to illustrate the oscillation of reaction potential (. a E s ).
See text for description of experimental procedure. (From an unpublished
study performed in the author’s laboratory by Ruth Hays and Charles B.
Woodbury.)

tion of pressures shown in the upper portion of Figure 67. There
the solid circles represent the pressures which were followed by food
reinforcement, and the hollow circles represent the pressures which
were too weak to deliver the food pellet. This distribution shows

30 6 PRINCIPLES OF BEHAVIOR

the phenomenon of intensity variability of a single organism within
a short period of time under practically identical external stimu-
lating conditions; the maximum pressure was more than three times
as great as the weakest, and about twice as great as the minimum
required to yield food pellet delivery. The upper portion of Fig-
ure 67 also suggests that the distribution of oscillating reaction
intensities conforms approximately to the normal “law” of chance.

Further illustration of the same general tendencies is presented
by the lower distribution in Figure 67. This shows the variability
of pressures by the same animal in the same apparatus after four
more days of 100 reinforcements each with an apparatus adjust-
ment which required the animal to make a pressure of 38 grams
before the food pellet would be delivered. Here we find an even
wider range of variability in pressure intensities than that shown
in the upper distribution. Again there is a general tendency for
the distribution to conform to the normal “law” of chance, though
in this case the conformity is not so close as in the upper distribu-
tion. Several other animals trained like the one whose record has
been given above, showed exactly similar tendencies in all respects.
Moreover, Hill (S), in connection with an investigation directed
at an entirely different objective, secured results on the conditioned
eyelid reaction which incidentally showed that both amplitude and
reaction latency under relatively constant conditions display an
oscillation closely comparable to that revealed by the rat pressures
of Figure 67. 1

Tables published by Thorndike ( 10 ) based on a line-drawing
experiment where learning apparently was effectively precluded by
human subjects whose eyes were closed, substantiate the results
yielded by Hays and Woodbury’s rats. Thorndike instructed his
subjects to draw what seemed to them to be a two-inch line, for
example. One subject drew 1,697 of these lines. This distribution
shows great variability or oscillation, with the majority of the
lengths falling in the central region much as shown in Figure 67.
Thorndike’s distribution revealed, however, a long, thin, asym-
metrical tail on the side of the longer lines. Since the rat data
mentioned above give no suggestion of this asymmetry, one may
reasonably conjecture that it was produced by some factor in the
experimental situation other than the primitive oscillation ten-
dency. A further examination of Thorndike’s tables reveals one

1 Received in a private communication and reported here with Dr. Hill a
permission-

BEHAVIORAL OSCILLATION 307

subject who drew a large number each of two-inch, four-inch, and
six-inch lines. These comparable distributions indicate a progres-
sive increase in the range of oscillation in the length of line drawn.
Additional study shows that the range of oscillation in this series
has an approximately linear relationship to the central tendency
of the lines actually drawn. Here again there is essential agree-
ment with the rat behavior shown in Figure 67.

A striking, though indirect, indication of the oscillation or

Fio. 68. Graphic representation of the empirical per cent of successful
reactions at the presentation of the cue syllable in the case of the learning of
205 nonsense syllables in series. The complete learning of each of the
syllables in question was preceded by six failures of reaction evocation (with
reinforcement), i.e., six presentations of the cue syllable which were not
followed by the correct response. (Reproduced from Mathematico-Deductive
Theory of Rote Learning, 5, p. 162.)

variability of behavior potentiality is presented in the early stages
of most simple conditioning situations where the experiment is so
set up that conditioned and unconditioned reactions are clearly
distinguishable. Under these circumstances it is common for the
conditioned stimulus to evoke its reaction on one occasion, yet not
on the next, in spite of the fact that because of intervening rein-
forcement the effective habit strength must have been stronger at
the time the failure occurred than it was at the time of the preced-
ing success (4) .

Notwithstanding the seemingly fortuitous occurrence of reac-
tions during the early stages in the acquisition of receptor-effector

PRINCIPLES OF BEHAVIOR

308

connections, the probability of evocation actually increases con-
tinuously with the increase of the effective habit strength and,
since drive is presumably constant, of effective reaction potential
(s#r). This may be shown by pooling the evocation results medi-
ated by a large number of a H R connections 'which increase in
strength at approximately the same rate. There is revealed by
this procedure not only a progressive increase in the probability
of reaction evocation but a characteristic sigmoid curve of increase.
The outcome of such a procedure is shown in Figure 68. The nor-
mal chance distribution of Figure 67 in the learning situation just
described would produce exactly the sigmoid learning curve seen in
Figure 68.

The above results taken as a whole indicate that the same
external stimulating conditions , operating through an approximately

identical habit structure ( a H R ) and a relatively constant drive ( D ),
and so an approximately constant effective reaction potential (bEr)>
will evoke distinctly diverse reactions .

ASYNCHRONISM OF BEHAVIORAL OSCILLATION

The phenomena represented in Figure 68 are presumably pro-
duced by the gradual rise of the effective strength of a habit above
the reaction threshold. In such a situation, if the JSr chances to
oscillate so as to be above the reaction threshold at the moment
of stimulation, overt reaction will occur; if it chances to fall below
the reaction threshold, the reaction will not occur. Reaction will
follow every stimulation only on condition that the momentary
effective reaction potential exceeds the reaction threshold by an
amount greater than the range of oscillation below it.

There is a related case in which the problem is concerned not
with the occurrence or non-occurrence of a reaction but with which
one of two or more competing incompatible reactions 1 will be
evoked by a given stimulus situation when the effective reaction
potential of each exceeds the reaction threshold by an amount
greater than the range of oscillation below it. Under such condi-
tions each reaction tendency, if not interfered with by another
incompatible one originating in the same stimulus situation, would
mediate its reaction at every stimulation. Suppose, now, that to
these conditions there be added a further one, namely, that the

1 Incompatible reactions are those which cannot be executed at the same
time.

BEHAVIORAL OSCILLATION 309

habit strength of one of two competing reaction tendencies exceeds
that of the other, but by an amount less than the range of oscilla-
tion below it. It follows that if the competing effective reaction
tendencies both oscillate upward or downward at the same time,
i.e., in synchronism, the one with the strongest habit strength will
always dominate, its reaction occurring at every stimulation and
the other reaction not occurring at all. If, however, the oscillations
of the two reaction tendencies are asynchronous or at least are not
perfectly correlated, then there may be expected an irregular alter-
nation, the relative frequency of the occurrence of each reaction
being an increasing function of the difference between the respective
habit strengths (see Table 3 and Figure 36, pp. 147 and 150).

Experiment reveals the latter state of things rather than the
former. It follows that in a trial-and-error learning situation of
this kind, where the strength of one of several competing reaction
tendencies is steadily increasing with respect to the others, domi-
nance by this reaction tendency would be attained gradually rather
than abruptly j this also is a fact. The outcome of such a process,
where the ultimately dominant habit was at the outset relatively
weak, is shown in Figure 24 (p. 108). It is abundantly clear that
the oscillation of effective habit strength is, to a considerable extent,
asynchronous.

POSSIBLE SUBMOLAR CAUSES OF BEHAVIORAL OSCILLATION

There is reason to believe that one of the ultimate physio-
logical or submolar causes of molar behavioral oscillation lies in
the variability in the molecular constituents of the nervous system,
the neurons. Blair and Erlanger have found, as the result of ex-
ceedingly delicate experiments on frog nerves, that neural response
thresholds and reaction latencies of individual axon fibers vary
spontaneously from instant to instant. They report:

When a preparation containing a fiber of outstanding irritability is
stimulated with shocks increased in strength by small steps from below
the fiber’s threshold, there are at first only rare responses. To elicit
a spike [electrical reaction] with every shock it usually is necessary to
increase the strength further by about 2 per cent.

They report further:

The time intervening between successive threshold shocks and the
resulting conducted axon spikes, or the shock-snike time, under exactly

PRINCIPLES OF BEHAVIOR

31°

comparable conditions is not constant. The range of fluctuation in the
case of the more irritable fibers may be as great as 0.5 tr, but usually is
about 0.2 to 0.3 <j; in the case of the less irritable fibers it may be more
than 2.4 o\ . . . (I, pp. 530-531.)

If to the spontaneous oscillation in irritability of the neural
conduction elements which mediate behavior, as reported by Blair
and Erlanger, there be added the random and spontaneous firing
of the individual neurons throughout the nervous system, which is
indicated by the experimental evidence reported by Weiss (see p.
45, above), there would seem to be ample grounds for expecting
oscillation to be a universal characteristic of organismic be-
havior {11).

The investigations just cited suggest not only the physiological
cause of behavior oscillation but its characteristic distribution.
Mathematicians have shown that the results of the joint action
of a multitude of independently varying small factors tend to
distribute themselves according to the so-called Gaussian or “nor-
mal law” of probability ( 2 ). Thus if 16 coins are tossed simul-
taneously a large number of times, and the number of heads com-
ing up at each toss is recorded, it will be found that the most com-
mon number obtained will tend to be 8 and that the frequency of
the other numbers of heads per throw tapers off symmetrically as
zero and 16 heads per throw are approached. In a similar manner
the spontaneous neural oscillations favoring a reaction stronger
than average may be thought of on the analogy of the heads of
the coins, whereas those favoring a reaction weaker than average
may be considered analogous to the tails of the coins. Accordingly
it is to be expected that in most cases the two opposing tendencies
will be about equal in number and so will approximately balance
each other, yielding a reaction of medium intensity. Occasionally,
however, a disproportionate number of neural phases favoring
strong or weak reactions will occur, just as a higher than average
number of heads or tails in the coin-tossing experiment sometimes
appears. On such occasions an unusually strong or an unusually
weak reaction will be made, e.g., an unusually long or an unusually
short line will be drawn. This, of course, agrees very well with
the facts of gross behavior variability as represented in Figure 67.
Other things equal, then, we may expect that the magnitude of the
contraction of each muscle involved in an act mediated by a recep-
tor-effector connection will vary as a function of the normal law
of probability.

This Table Shows (1) the Relative Frequencies (in Percentages) of the Probability Integral over Each One-Tenth
of the Standard Deviation, and (2) the Cumulative Percentage Values for the Same Integral. The Distribution of
Values Is Cut Off Arbitrarily at 2.5 a at Each Side of the Central Tendency. Note That One-Half of the Probability
Distribution Is Shown in Columns b and d , and That the Other Half Is Shown in Columns b ' and d '. It Will Be Observed
That Columns c and d Are Derived Directly from Columns a and b Respectively. (Adapted from Thorndike, 9.)

> *>5

& ? c 3 © . G a

H U © O

££*«•£ 130
° -2

49.37

63.35

67.29

61.18

64.91

68.51

71.94

75.17

78.18

80.96

83.50

85.80

87.86

89.69

91.29

92.68

93.88

94.90

95.76

96.48

97.08

97.57

97.97

98.29

98.54

98.74

100.00

Deviations
from a Point
at -2.5a from
the Central
Tendency of
Probability

b bbbbb bbbbb bbbbb bbbbb bbbbbb

*0 ©NCOOO •-« p CO xj« O OSCOO)0 rn CO -*< p CO 00 O •— «

cs ci n n ci c*i co co co co co co co co co ^ ^ ^ o* ^ ^ io

Percentage of
Probability
Lying Between
Each a Entry
and the One
Preceding It
in Column (o')

S SS$5E§ 5e3o®3 8 S 88 S SS 8 SS 8

co co co co co co cocdcoeici

Deviations
from the
Central
Tendency
of Probability

b bbbbb bbbbb bbbbb bbbbb bbbbb

O •-« C* CO ^ *0 ONCOOO »-« C* CO ^ O (ONCDOO ^<NC0^»0

• • • • • • ••••• ••••• ••• • f

+ + + + + ++ + + : . -----

Cumulative
Percentage
of Probability
Lying Between
—2.5 <r and
the Entry
in Column (a)

.00

.20

.45

.77

1.17

1.66

2.26

2.98

3.84

4.86

6.06

7.45

9.05

10.88

12.94

15.24

17.78

20 56

23 57

26.80

30.23

33.83

37.58

41.45

45.39

49.37

Deviations
from a Point
at -2.5 a from
the Central
Tendency of
Probability

b bbbbb bbbbb bbbbb bbbbb bbbbb

O -jNWTfiO C0b-C0 0>0 CONCOOiO OQ CO iO

h «hhh hhhhci cicicicici

Percentage
of Probability
Lying Between
Each c 7 Entry
and the One
Preceding It
in Column (a)

S 33S3SS 8 K 888 833§g SSSS33 g^SoSg

HH F-4 V-4 r-« oi c4 (N CN CO CO CO CO CO CO CO CO

Deviations
from the
Central
Tendency
of Probability

"5*

b bbbbb bbbbb bbbbb bbbbb bbbbb

*0 -^COC^^p a> CO o lO CO <N O 0> ^CONmO

1 1 1 II 1 1 1 1 II 1 1 1 1 1 1 " ‘ 1 ' 11,1

311

PRINCIPLES OF BEHAVIOR

3 1 2

SOME FURTHER CONSEQUENCES OF THE SIMUL/TANEOUS
FORTUITOUS VARIATION OF A VERY LARGE NUMBER

OF INDEPENDENT FACTORS

Mathematicians ( 2 , p. 33 ff.) have determined by appropriate
methods the outcome of what amounts to a coin-tossing experiment
in which the number of coins is infinite and an infinite number
of throws are made. The results of this mathematical procedure
are conveniently presented in the form of a table, which is ex-

DEVIATIONS FROM CENTRAL TENDENCY IN a UNITS

Fiq. 69. Graphic representation of the distribution of normal probability
to which the intensities of reactions evoked by repetitions of the same stimu-
lus are believed to approach as a first approximation. This figure has been
plotted mainly from columns a, b, a, and 6' of Table 9. Note the bell-shaped
contour of the distribution.

tremely useful for reference, as it gives the standard form of dis-
tribution toward which all situations involving the action of nu-
merous chance factors approach. An abbreviated adaptation of
such an assemblage of theoretical chance values is shown in Table 9.

A graphical representation of one aspect of this particular
phase of probability, that of the chance that the joint action of
the factors will deviate by a given amount from the central ten-
dency, is given in Figure 69. This figure was derived from Table 9
by plotting the values in columns b and b' as a function of the
values in columns a and a'. Figure 69 should be compared with
empirical Figure 67. It will be noted that Figure 69 has a smooth

BEHAVIORAL OSCILLATION 3 1 3

contour, whereas Figure 67 varies by coarse steps; this difference
is due to the infinite number of infinitesimal factors upon which
Figure 69 is based.

The graphical representation of a second aspect of the joint
action of an infinite number of small independent chance factors
is given in Figure 70. This figure was also derived from Table 9,
by plotting the values in col-
umns d and d ' as a function
of the values in columns c
and c'. Figure 70 should be
compared with the empirical
graph shown in Figure 68. It
will be noticed that Figure
68 resembles Figure 70 in
that it is markedly sigmoid
in shape; this fact strongly
supports the hypothesis that
a distribution of many inde-
pendent chance factors pro-
duced the oscillation which
occurred during the learning
of the nonsense syllables,
from the results of which
Figure 68 was derived. Fig-
ure 68 differs markedly from
Figure 70 in that the upper
portion of the S-shaped
curve is much more extended than is the lower portion. This
presumably is the result of the slower rate of habit strength acqui-
sition as it approaches its physiological limit (see p. 116).

THE MOLAR CONCEPTS OP BEHAVIORAL OSCILLATION ( bOr ) AND OF
MOMENTARY EFFECTIVE REACTION POTENTIAL (bEr)

At this point of our analysis we may formulate the molar con-
cept of behavioral oscillation. On the basis of the submolar con-
siderations presented above it is believed that the variability of
reaction under seemingly constant conditions is due to the action
of an oscillatory force upon the effective reaction potential ( B E R ),
This oscillatory force will be represented by the symbol B 0 R . The
momentary state of bE r under the influence of B 0 R will be called

DEVIATION FROM 2.5 o PROBABILITY

Fio. 70. Graphic representation of the
cumulative per cent of normal probability
from — 2.5 a to +2.5 a. This figure has
been plotted from columns c, d, c, and df
of Table 9. Because of its shape, this
representation of the probability function
is called the ogive.

3 ! 4

PRINCIPLES OF BEHAVIOR

the momentary effective reaction potential and will be represented
by the symbol a E R .

The evidence at present available indicates that a O R rather
closely approaches a normal chance distribution. A number of
other critical matters in the situation are much less clearly evident.
Among these latter problems is the question: Does the range of
oscillation vary for a given organism, and if so upon what does
this variability depend? This problem is particularly acute as
the value of B E R rises from zero to the reaction threshold. Closely
related is the question of the direction of the momentary shift of
bE r under the influence of a O R . For example, the action of a O R
might be wholly positive, causing a E R always to oscillate upward;
or its action might be distributed equally in both the positive and
the negative direction. The lower portion of Figure 67 together
with the conditions and facts of the acquisition of skill, which is
believed to be closely related to the situation which yielded that
distribution, suggests that a W R may oscillate both upward and
downward. On the other hand, the shape of certain curves of
simple trial-and-error learning suggests that the action of a O R on
B E R may be wholly negative and that its range may be substan-
tially constant.

Partly as a means of facilitating exposition, but partly also as
a means of opening the associated problems to much needed inves-
tigation, both empirical and theoretical, it has been decided to
try out here the conceptually rather simple hypothesis suggested
by the curve of simple trial-and-error learning, namely that B 0 R
is an oscillating inhibitory potentiality , that it acts against effective
reaction potential { a EJ R ), that the distribution of a O R conforms to
the normal law of chance , that the mean value of a O R and its range
are both constant , and that the action of a O R on the a E R as applied
to the several individual muscles is non- correlated,

THE RESOLUTION OF THE “REACTION-EVOCATION” PARADOX

At this point we find ourselves in possession of principles ade-
quate to explain the reaction-evocation paradox; i.e., how a stim-
ulus may evoke a reaction which has never been conditioned either
to it or to a stimulus in its stimulus-generalization range. The
problem may, for convenience of exposition, be divided into two
parts, (1) the resolution of the reaction-evocation paradox as ap-

BEHAVIORAL OSCILLATION

3 *5

plied to muscular contraction purely as such (to the so-called
actones of Murray, 6, p. 54 ff.), and (2) the resolution of the same
paradox as applied to acts , i.e., muscular contractions from the
point of view of their effects upon the environment, particularly
as these bear on the subsequent reinforcement of the movements
in question.

The ultimate effector molar unit in habitual action is believed
to be the individual muscle. This means not only that the action
of each muscle in every coordinated movement must be mediated
by a separate habit, but that every momentary phase of the con-
traction of every such muscle (since its proprioceptive cues are
constantly changing) must be mediated by what is in some sense
a different habit. Viewed in this manner, the contraction intensity
of a given muscle, as mediated by the results of a given reinforce-
ment, is the summation of the action of an uninterrupted chain or
flux of habit, each phase of which is more or less distorted by the
oscillation function. In this connection it must be recalled (p. 308)
that the strength of each habit oscillates largely independently of
all the others.

Now, nearly all movements are mediated by the coordination
of sizable muscle groups. If the contraction of one muscle of such
a group should vary in its intensity, that of the others remaining
constant, the joint movement produced by the group as a whole
will inevitably deviate in one respect or dimension from what it
otherwise would have been. Since the contraction of each muscle
is mediated by distinct habits, the contraction of all the muscles
of a group will oscillate independently. Thus coordinated move-
ment as such may be said to have as many dimensions of variation
as there are muscles involved in its production. It follows from
these considerations that infinitely varied movements other than
those involved in the original conditioning process will inevitably
be evoked by the impact of the conditioned stimulus, S. In this
way the reaction-evocation paradox, from the point of view of
movement as such, finds its resolution.

From the point of view of action as defined in the first para-
graph of this section, it may be pointed out that behavioral oscil-
lation gives rise to qualitative as well as quantitative differences.
For example, if a particular muscle in a group mediating the strik-
ing of a typewriter key acts too weakly, the impression may be too
faint to be legible; on the other hand, if some other muscle oscil-
lates in the direction of too strong a contraction, the stroke may be

3 1 6 PRINCIPLES OF BEHAVIOR

diverted to one side and a quite different key will be hit. In such
a case a qualitatively different act or outcome may be said to have
resulted from a quantitative deviation in an act one or movement. 1
Thus it is clear that very varied acts may result from the rein-
forcement of an extremely narrow zone of movements. In this
way there emerges from the analysis the substance of what may be
called response intensity generalization. Hence the reaction-evo-
cation paradox, from the point of view of action and goal attain-
ment, finds its resolution.

THE INFLUENCE OF THE OSCILLATION PRINCIPLE ON THE STATUS
AND METHODOLOGIES OF THE BEHAVIOR SCIENCES

It is quite clear from the foregoing that the concrete manifes-
tation of empirical laws (such as those concerning the acquisition of
habit strength, p. 102 ff.) is bound to be greatly blurred by be-
havioral oscillation. Indeed, at first sight it might be thought

that behavioral oscillation would preclude the possibility of any

_ •

exact behavior science whatever. As a matter of fact, this pessi-
mistic view is seriously held in certain quarters.

It must be confessed that behavioral oscillation does impose a
grave handicap on all the social sciences; generally speaking, it
precludes the possibility of deductively predicting the exact mo-
mentary behavior of single organisms. However, with an intimate
knowledge of the history of the organism in question and a good
understanding of the molar laws of behavior, it should be possible
to predict within the limits imposed by the oscillation factor what
the subject will do under given conditions. That behavior pre-
diction has this limitation may be disappointing to some, particu-
larly to individuals engaged in clinical practice, but there seems
no escape from this difficulty; our task as scientists is to report
what we find, rather than what we or our friends might wish the
situation to be.

From the point of view of the general molar laws of behavior,
the situation is far more satisfactory. Because of the tendency of
large numbers of independent chance factors to distribute their
influence more or less symmetrically (Figure 69) according to the
Gaussian or normal law, it comes about that by various rather

1 Fortunately, in most life situations reinforcement will follow about
equally well movements possessing a considerable range of variability. Were
this not so, organisms a 8 now constituted could hardly survive.

BEHAVIORAL OSCILLATION

3 1 ?

simple statistical devices it is possible, when many comparable
measurements have been made on the same individual’s behavior
or on that of a large number of comparable individuals, to isolate
the central tendencies from the measures of individual reactions,
more or less distorted as they are by the oscillation factor, and thus
to reveal a close approximation to the laws which are operating.
Mathematicians have shown that, other things constant, the dis-
tortions due to such chance factors vary inversely with the square
root of the number of observations from which the central tendency
is calculated. Thus the deviation of a mean calculated from 64
measures will in general differ from the “true” mean, i.e., that which
might be calculated from an infinite number of comparable meas-
ures, by only half as much as that calculated from sixteen meas-
ures; the square root of 64 is 8, the square root of 16 is 4; and 4
is half as great as 8.

On the above principle it is evident that complete absence of
blurring of the mean, due to oscillation and other chance irrelevant
factors in the situation, is attained only when the number of meas-
ures becomes infinite; this means that absolutely exact empirical
laws are never attainable. It follows that the most that can be
hoped for in the empirical checking of the implications of be-
havioral laws must be greater or less degrees of approximation;
and even this approximation can be attained only at the cost of
great care and vast labor in the massing of data. Indeed, the uni-
versality of oscillation in organismic behavior is the main reason
why the social sciences have been forced so extensively to employ
statistical methods.

Finally, it may be said that the principle of behavioral oscil-
lation is to a large extent responsible for the relatively backward
condition of the social, as compared with the physical, sciences.

SUMMARY

Variability, inconsistency, and specific unpredictability of reac-
tion under seemingly constant conditions are universal character-
istics of the molar behavior of organisms, attested alike by general
observation and by quantitative experiment. Neuro-physiological
investigations suggest that behavioral oscillation arises from the
spontaneously variable action of an enormous number of small
factors (nerve cells), each acting independently to increase or
decrease the intensity of reactions mediated by receptor-effector

PRINCIPLES OF BEHAVIOR

3 l8

connections. Where learned behavior is concerned, the oscillation
is presumably of the effective reaction potentiality ( a E R ). In gen-
eral confirmation of this view, typical experimental determinations
show that with the effective habit strength and primary drive sub-
stantially constant, behavior evoked by successive repetitions of
the same stimulus presents a close approximation to a normal
probability curve. Moreover, evidence derived from trial-and-error
learning situations demonstrates in a convincing manner that the
oscillation associated with each habit tendency is largely, if not
totally, uncorrelated with that of the others, i.e., that the oscilla-
tions of different effective habit tendencies are essentially asyn-
chronous.

Additional indirect confirmation of the general Gaussian dis-
tribution of the oscillation function is found in the sigmoid shape
of certain learning curves when plotted in terms of the per cent
(or probability) of reaction evocation. The characteristically more
protracted upper portion of these curves is presumably due in the
main, at least, to the progressively slower rate of habit-strength
acquisition as the physiological limit is approached.

Two important implications of the oscillation principle may be
noted. The first yields an explanation for the superficial paradox
that a stimulus is able to evoke reactions which are more or less
distinct from any ever conditioned to it or to any other stimulus
on the same stimulus continuum. The explanation lies in the ten-
dency of the oscillation function to modify the intensity of every
muscular contraction involved in every coordinated reaction; this
makes the act evoked more or less different from any act involved
in the original reinforcement. This amounts in effect to response
intensity generalization.

A second implication of the principle of oscillation is that no
theory of organismic behavior can ever be expected to mediate
the precise prediction of the specific behavior of any organism at
a given instant. However, because of the general regularity of the
probability distribution in both its symmetrical and skewed forms,
it will always be possible to predict approximately the central ten-
dencies of behavior data from either individual organisms or groups
of organisms which are under the influence of approximately the
same antecedent factors and which share substantially the same
neural and receptor-effector equipment. The oscillation factor ex-
plains why all of the behavior sciences derive their empirical laws
from averages, and why their quantitative investigations necessi-

BEHAVIORAL OSCILLATION

319

tate the securing of such great numbers of data. Finally, be-
havioral oscillation is believed to be, to a considerable extent,
responsible for the present relatively backward state of the be-
havior (social) sciences.

On the basis of the considerations put forward in the preced-
ing pages, we now formulate our tenth primary molar principle:

POSTULATE 10

Associated with every reaction potential ( sEr ) there exists an inhibi-
tory potentiality ( sOr ) which oscillates in amount from instant to instant
according to the normal “law” of chance. The amount of this inhibitory
potentiality associated with the several habits of a given organism at a
particular instant is uncorrelated, and the amount of diminution in s&r
from the action of sOr is limited only by the amount of sEr at the time
available.

From Postulate 10 there follows Major Corollary III:

MAJOR COROLLARY m

Each muscular contraction involved in any increment of habit tendency
( A sHr) oscillates from instant to instant in the reaction-intensity potenti-
ality which it mediates, thus producing a kind of response generalization
in both directions from the response intensity originally reinforced.

NOTES

Mathematical Statement of Postulate 10
The mathematical statement'of Postulate 10 is given by the following equation

A — b^r — bO'r, (44)

where,

bEh — the momentary effective strength of a reaction potential as modi-
fied by the oscillatory potentiality, gOst

where,

flO's = eOr when bEr ^ bOr,

and,

sO'r — bEr when bEr < sOr,

where,

zero ^ bOr ^ 6 <r, a being a constant,

and,

the probability (p) that bOr takes on values between zero and 6 <r is p,

where,

(«- 3 «)*
2 **

320

PRINCIPLES OF BEHAVIOR

On the Oscillation of Effective Reaction Potentiality

A conception of behavior oscillation much like the one here elaborated was
published by the author many years ago (4). Spearman ( 8 , p. 323) incorporated
it into his system of group factors which are supposed to determine human be-
havior, but otherwise the idea seems not, as yet, to have found acceptance or
utilization. It would seem that as the theory of behavior grows more adequate
and general, this principle in some form must find explicit recognition.

The equivalent of the principle of behavior oscillation appears as Postulate 15
of M at hematico- Deductive Theory of Rote Learning (5, p. 74). In that postulate,
however, it was the reaction threshold which was supposed to oscillate, whereas
in the present system it is reaction potential ( sE R ) which is postulated as varying.

Finally, in the present system the principle of oscillation appears to be shifting
from the status of a postulate or primary principle to that of a theorem or second-
ary principle. Such a change in the status of the principle of oscillation would be,
of course, in accordance with the principle of parsimony, which is to the effect
that, other things equal, the number of assumptions should be as few as possible.

The Derivation of Table 9 and Figures 69 and 70

Table 9 and Figures 69 and 70 are derived from the equation,

2a* 9

where N is the total number of chances (population) involved, y is approximately
the number of these chances falling within a given interval, S is the extent of that
interval in a units, x is the distance of the midpoint of that interval from the
central point of the distribution of chances, a is the standard deviation of the
distribution of chances, and ir and e are mathematical constants with approximate
values of 3.1416 and 2.718 respectively (7, p. 13).

The method of deriving columns b and b' of Table 9 (from which columns d
and d f are derived) is illustrated by the following example :

Problem’. To calculate the population of probability or chances falling within
a range of S = .1 a (from x — .05 <r to x + .05 <r) the midpoint of which is located
at a distance, x = 1.35 <r, from the central tendency of the distribution of chances,
the total population of chances being N = 100, the standard deviation of the
distribution of chances being the unit of measurement of the range of the chances,
i.e., <r = 1. Substituting these several values in the equation, we have,

-1.35*

100 X .1
V2 X 3.1416

2X1*

X 2.718

in —1-8225

= -y . - . X 2.718 2

y/ 6.2832

' 10

s X 2 718~- MUS
2.506 X

= 3.9893 X .40248
= 1.6056,

BEHAVIORAL OSCILLATION 321

which agrees to the second decimal place with the entry in column b, opposite
that of 1.3 a in column a in Table 9. This entry covers the range from 1.4 a to
that of 1.3 <r, the midpoint of which is 1.35, which is taken as the value of x in
the above computations.

Behavioral Variability as Caused by the Introduction of Unaccustomed
Components into the Conditioned Stimulus Compound

It is probable that an appreciable portion of the gross variability in the re-
sponses of organisms under approximately constant conditions of habit organi-
zation arises from the intrusion of a stimulus component, not present when the
conditioning originally occurred, into the stimulus complex normally evoking
the reaction. By the principle of afferent neural interaction (p. 42), such an
intrusion would change to a certain extent the afferent impulse produced by the
originally conditioned stimulus components. This, by the principle of the
generalization gradient (p. 185), should weaken the resulting reaction, thus
producing an oscillation in a downward direction. Numerous other variants in
the stimulus would produce analogous quantitative variations in action evocation.

REFERENCES

1. Blair, E. A., and Erlanger, J. A comparison of the characteristics of

axons through their individual electric responses. Amer. J. Physiol ..
1933, 106, 524-564.

2. Brown, W., and Thomson, G. H. The essentials of mental measurement .

London: Cambridge Univ. Press, 1921.

3. Hill, C. J. Retroactive inhibition in conditioned response learning.

PhJ). thesis, 1941, on file Yale Univ. Library.

4. Hull, C. L. The formation and retention of associations among the in-

sane. Amer. J. Psychol., 1917, 28, 419^35.

5. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T.,

Fitch, F. B. Mathematico-deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

6 . Murray, H. A. Explorations in personality. New York: Oxford Univ.

Press, 1938.

7. Rietz, H. L. Handbook of mathematical statistics. New York: Hough-

ton Mifflin Co., 1924.

8 . Spearman, C. The abilities of man. New York: Macmillan Co., 1927.

9. Thorndike, E. L. Theory of mental and social measurements. New

York: Teachers College, Columbia Univ., 1916.

10. Thorndike, E. L. The fundamentals of learning. New York: Teachers

College, Columbia Univ., 1932.

11. Weiss, P. Functional properties of isolated spinal cord grafts in larval

amphibians. Proc. Soc . Exper. Biology and Medicine, 1940, U, 350 £f.

CHAPTER XVIII

The Reaction Threshold and Response Evocation

It will be recalled that more than once in the preceding pages,
when discussing symbolic constructs such as sH R , D, bEr, %Rj an( ^
bBTr, we have emphasized the scientific hazards involved in their
use. In this connection it has always been made clear that these
dangers can be obviated only by having the constructs securely
anchored in two temporal directions: (1) in objectively observable
and measurable antecedent conditions or events, and (2) in objec-
tively observable and measurable consequent conditions or events.
Up to the present we have satisfied these requirements reasonably
well in respect to the first or antecedent direction by laying down
the conditions which culminate in effective reaction potential (bEr).
To this end there have been shown in succession (1) the conditions,
and (2) the relevant principles which generate habit strength
( g H R ) ; which determine generalized habit strength ( sHr ) > which
show how drive ( D ) is generated, how habit strength and drive
combine to produce reaction potential (sE R ) f how inhibitory poten-
tials ( I R and 8 Ir ) are generated, how these combine with reaction
potential to produce effective reaction potential (bEr), and how
oscillation ( s O«) combines with a E R to produce momentary effective

reaction potential ( bEr )•

With the critical construct 8 E R thus securely anchored on the
antecedent side, we are at last free to consider the events an
principles whereby it is anchored on the consequent side. In
general the latter relationships are somewhat simpler than the
former. Briefly stated, the consequent anchoring events are reac-
tions ( R ), i.e., the movements or other activities of the organism.
For the most part these reactions are susceptible of direct observa-
tion and automatic objective recording.

At present the clearest and most dependable single quantitative
relationship subsisting between B E R and R seems to be that of the
probability ( p ) of the occurrence of the response following stimu-
lation. Supplementing the probability-of-reaction-evocation rela-
tionship ( p ) are three additional functional relationships which
materially contribute to the anchoring of the b Er construct on the
consequent side. These are: the latency of the reaction, the resist-

322

REACTION THRESHOLD-RESPONSE EVOCATION 323

ance of the reaction potential to experimental extinction, and the
amplitude of the reaction. In the delineation of the relationship
of 8 E r to R we shall accordingly take up first that based on the
probability of reaction evocation. As a preliminary to this, how-
ever, it will be necessary to introduce the concept of the reaction
threshold.

THE CONCEPT OF THE REACTION THRESHOLD ( S L R )

As used in neurophysiological, psychological, and behavior the-
ory and empirical practice, the term threshold implies in general a
quantum of resistance or inertia which must be overcome by an
opposing force before the latter can pass over into action. So
defined, the threshold concept fits many natural situations to which
it is not customarily applied. Thus in beginning to drag a heavy
object over a surface, one must often apply many pounds of trac-
tion before the weight begins to move perceptibly; if traction is
gradually increased, there comes a point at which the addition of
one more ounce to the pull starts the weight moving. The traction
at this point would be the approximate threshold.

In an analogous manner, an appreciable weight must be placed
on the skin before the subject can report its presence with a given
degree of consistency ; a certain amplitude of air vibration must

reach the ear before the subject can consistently report a sound; a
certain intensity of light must enter the eye before the subject
can consistently report a color; the forearm must be moved a cer-
tain minimal number of units of arc at its joint before the subject
can consistently report that passive movement has occurred. All
of these are stimulus thresholds, traditionally supposed to be based
primarily on the resistance or inertia of the receptor mechanisms.

In the early days of experimental psychology, as a result of
the activities of Weber and Fechner (I), much time and energy
were devoted to the determination of such thresholds. Because the
early experimental psychologists were chiefly German and because
of the prevalence of certain philosophical beliefs in Germany at the
time, particularly those associated with metaphysical idealism,
these minimal reportable stimulations were thought to represent
a ^ Quantitati v © relationship between the physical and the psychic;
this supposed transition of the physical into the psychic was thus
conceived as a process of the physical stimulus entering the door
of consciousness, hence the use of the word threshold . Accordingly,

324 PRINCIPLES OF BEHAVIOR

the Latin word limen (threshold) is still commonly used as a
synonym for the threshold in psychophysics. For this reason the
threshold in psychology is represented by the symbol L\ this sym-
bol, with the addition of a pair of qualifying subscripts, S and R ,
is employed in the present work to represent the reaction threshold,
thus: s L r .

Proceeding to the matter of the quantitative operational defini-
tion of the threshold, it may be pointed out that in the chronaxie
determinations of neurophysiology the threshold is defined as that
minimal electrical current acting on any irritable tissue for an
indefinite period which will evoke detectable activity, e.g., a dis-
charge along a nerve fiber or a movement in a bit of muscle ( 2 , p.
78 ff.). In an analogous manner, the reaction threshold (bLr) is
defined as the minimal effective reaction potential iaE R ) which will
evoke observable reaction; i.e., no reaction will occur unless

sEr — sEr

is greater than zero. This difference we shall call the super-
threshold effective reaction potential .

INDIRECT DEMONSTRATIONS OF THE EMPIRICAL REACTION

THRESHOLD

In ordinary behavior the reality of an empirical reaction thresh-
old is demonstrated perhaps most clearly by the fact that in con-
ditioning and other learning situations several reinforcements are
frequently required before the stimulus will evoke the reaction.
For example, in the conditioning of lid closure the conditioned reac-
tion may be recorded as quite distinct from the blink which is
associated with the air puff or whatever the reinforcing stimulus
happens to be. It thus comes about that the question of whether
or not the conditioned stimulus is able to evoke the reaction to
which it is being conditioned, is readily determined empirically at
all stages of the learning process without the usual complication
of extinction effects. Utilizing this circumstance, Hill ( 5 ) found
in one experiment that 15 per cent of his subjects showed their first
conditioned reaction at the second reinforcing stimulation, 16.7 per
cent at the third, 11.7 per cent at the fourth, 8.3 per cent at the
fifth, 6.7 per cent at the sixth, and 3.3 per cent at the seventh stim-
ulation. Other things equal, the number of reinforcements required
before the first reaction evocation is a quantitative indication of

REACTION THRESHOLD-RESPONSE EVOCATION 325

the height of the reaction threshold. Thus those subjects giving
their first conditioned reaction at the sixth or seventh reinforce-
ment presumably had higher empirical reaction thresholds than
did those groups which gave their first conditioned reaction at the
second or third reinforcement.

A rather different quantitative illustration of the empirical
reaction threshold is presented in Figure 50 (p. 228). A careful
examination of this figure shows that both fitted curves originate
at the same point, namely, at a value of four extinction increments
(A Ir) below the point at which response would be evoked, i.e.,
below the empirical reaction threshold. This seemingly paradoxical
result is distinctly revealing as to the nature of the reaction thresh-
old; since, as we have seen, reactions will not occur if the excita-
tory potential is below the reaction threshold, this extinction meas-
ure of effective excitatory potential cannot function when 8 ^r is
less than 8 Lr. This means that the zero value of any response
scale falls exactly at the reaction threshold.

Nevertheless we naturally wish to know how far the reaction
threshold is above the absolute zero of effective reaction potential,
or just no b 'E r at all. Even though the extinction-reaction tech-
nique cannot directly enter this subthreshold region, it is possible
to determine the shape of the learning function at numerous other
points which are well above the reaction threshold, as shown for
example by the circles in Figure 50. The curve or law of the
function so determined, when extrapolated backward from the
points empirically established to where N = 0 (and so, presumably,
b'Er — 0) , yields an approximation to a value impossible of direct
measurement. Thus the mean empirical reaction threshold of rats
under the circumstances of Perin’s experiment (5) purports to be
an excitatory potential which would be approximately neutralized
by the first four extinction reactions of the test series.

Before leaving this subject it must be pointed out that while
the empirical reaction threshold includes the true or inertial thresh-
old ( b Lr ), the two are not identical. There is good reason to
believe (see sixth terminal note) that not only the two types of
empirical reaction thresholds just described, but all minimal stim~
ulus thresholds in psychophysics are the sum of the true inertial
threshold plus an artifact of undetermined magnitude which arises
from the action of the oscillation function ( a O R ). As yet no attempt
has been made to determine the relative magnitude of the two
factors entering into either the empirical reaction threshold or the

PRINCIPLES OF BEHAVIOR

3 26

stimulus threshold of psychophysics; such a determination, while
complicated and necessarily indirect, should be possible; when made
it would not be surprising if the oscillation component is found to
exceed in magnitude the true or inertial reaction threshold (sLr).
For the purposes of the present preliminary analysis, the quantita-
tive separation of the two presumptive components is not neces-
sary. 1

THE FUNCTIONAL RELATIONSHIP OF REACTION-EVOCATION

PROBABILITY ( p ) TO THE EFFECTIVE
REACTION POTENTIAL (sEr)

It will be recalled that as the result of certain considerations
put forward in Chapter VIII (p. 102 £f.) we concluded that habit
strength is a simple positive growth function of the number of
reinforcements. Since reaction potential is a joint multiplicative
function of habit strength and drive (p. 242), it follows that so
long as drive remains constant, reaction potential will also closely
approximate a simple growth function of the number of reinforce-
ments. In the case of some types of response, such as salivary
secretion and the galvanic skin reaction, the mean amplitude of
response in simple learning situations has been found to be a posi-
tive growth function of the number of reinforcements (p. 103 ff.) ;
this suggests a very simple (linear) relationship between the am-
plitude of such learned responses and effective reaction potential
(- bEr) .

Many reactions, such as the bar-pressing movements of Perin’s
rats, approach the all-or-none type, which differs appreciably from
the galvanic skin reaction, salivary secretion, etc. The all-or-none
type of reaction introduces into the situation (1) the reaction
threshold and (2) the oscillation of habit strength (p. 304 ff.).
Owing to the simplicity of the threshold concept and to the fact
that the oscillation function has already been fairly well established
by independent investigations, the natural and strategic way to
conceive the progress of learning from the response side is found
in the probability (p) that the impact of the conditioned stimulus
will evoke the response. Accordingly our examination of the quan-
titative relationship of effective reaction potential to the four
modes of its observable manifestation in action will begin with the

1 For further elaboration of this point, see sixth terminal note.

REACTION THRESHOLD-RESPONSE EVOCATION 327

probability of the reaction evocation of the all-or-none type of
response.

Let it be assumed that the habit strength of an all-or-none
type of reaction is reinforced 36 times, with uniform time inter-
vals between reinforcements great enough to prevent the accumu-
lation of appreciable amounts of reactive inhibition ( I R ). For
the sake of simplicity in theoretical calculation, let it further be
assumed that the drive is constant, that this and the conditions
of reinforcement are such that the asymptote of effective reaction

0 2 4 6 8 10 12 M 16 18 20 22 24 26 2 8 30 32 34 35

ORDINAL NUMBER OF REINFORCEMENTS (N)

Flo. 71. Diagram showing the gradual movement of the zone of reaction-
potential oscillation (up-ended bell-shaped areas) across the reaction threshold
( sLr ). The upper or growth curve represents bRr as a function of N and is
plotted from columns 1 and 2 of Table 10. For further explanation see text.

potential (, gE R ) with unlimited practice will be 80 wats (p. 134 ff.),
and that the nature of the reinforcing agent and related conditions
surrounding the reinforcement process is such that the increment
of effective reaction potential (aA) at each reinforcement will
be approximately one-twentieth of the difference between the effec-
tive reaction potential just preceding that reinforcement and the
80-wat asymptote. The reaction potentials just preceding each
stimulation (and reinforcement) have been calculated on the above
principle and are presented in numerical detail in the second column
of Table 10. These values are represented graphically by the
upward-arching growth curve shown in Figure 71.

328

PRINCIPLES OF BEHAVIOR

Now, since reaction can be evoked only when the effective reac-
tion potential exceeds the reaction threshold, i.e., when

sE r > sLr

and since by hypothesis in the present instance,

s L r = 10 wats,

it follows from the values in column 2 of Table 10 that the con-
ditioned stimulations associated with the first three reinforcements
cannot evoke the reaction being conditioned. Generalizing, we
arrive at our first corollary:

I. In the original learning of reactions of the all-or-none type,
at least one and often a number of reinforcements are required be-
fore the reaction can be evoked by the conditioned stimulus alone,
the number of reinforcements before reaction evocation being a
decreasing function of the steepness of slope of the learning curve.

It does not follow, however, that when

sE r > sL r

the impact of the conditioned stimulus will necessarily evoke the
reaction being conditioned. This uncertainty comes from a num-
ber of independent considerations. The one which especially con-
cerns us here is the oscillation principle, discussed at some length
in an earlier chapter (p. 304 ff.). In order to illustrate the opera-
tion of the oscillation principle in the present situation, let it be
assumed that the factors determining the oscillation of reaction
potential are such that when operating at their maximum they are
sufficient to neutralize any superthreshold effective reaction poten-
tial up to 50 wats which may be present, but when operating at a
minimum, their neutralizing effect would be zero. Moreover, in
accordance with considerations put forward in an earlier chapter
(p. 308 ff.), the magnitude of these depressing effects presumably
varies from moment to moment in a symmetrical manner about a
central tendency according to the Gaussian “law” of probability.
For convenience it will be assumed that the magnitude of oscillation
varies over a range of 5 o (standard deviations) ; thus the standard
deviation of oscillation (o 0 ) in the supposed situation would have
a value of 50 wats divided by 5, which yields a quotient of 10 wats.

With these values available and with the aid of a suitable table
(based on the assumption of an infinite sample) we may determine
the probability of reaction evocation after each reinforcement. The

REACTION THRESHOLD-RESPONSE EVOCATION

329

L TABLE 10

A Table Showing the Several Steps op the Derivation of the Prob-
ability of Reaction Evocation as a Joint Function of the Reaction
Threshold, the Strength of Effective Reaction Potential ( s Er) and
the Magnitude of the Oscillation of the Reaction Potential. Strictly
Speaking, These Values, Particularly Those of the Probability Function
( p), Presuppose an Unlimited Sample of Homogeneous Behavior.

Number of
Preceding
Reinforce-
ments

Effective

Reaction

Potential

(sEr)

Reaction

Potential

(sEr — S Lr)
III

(sEr —sLr)

<ro.

Where
(To = 10

Probability
of Reaction
Evocation (p)
Derived from
Table 9 and
Column IV

8
9

20
21
22

0.00

4.00

7.80

11.41
14.84
18.10
21.19
24.13

26.93
29.58
32.10

34.50
36.77

38.93
40.99

42.94
44.79
46.55
48.22

49.81
51.32
52.76
54.12

55.41
56.64

57.81

58.92

59.97

60.97

61.93
62.83
63.69

64.50
65.28
66.01
66.71
67.38

1.41

4.84

8.10

11.19

14.13

16.93
19.58
22.10

24.50
26.77

28.93
30.99

32.94
34.79
36.55
38.22

39.81
41.32
42.76
44.12
45.41
46.64

47.81

48.92

49.97

50.97

51.93
52.83
53.69

54.50
55.28
56.01
56.71
57.38

.141

.484

.810

1.119

1.413

1.693

1.958

2.210

2.450
2.677
2.893
3.099
3.294
3.479
3.655
3.822
3.981
4.132
4.276
4.412
4.541
4.664
4.781
4.892
4.997
5.097
5.193
5.283
5.369

5.450
5.528
5.601
5.671
5.738

1.66

3.84

7.45

12.94
20.56
30.23
37.45
49.37

57.29
64.91

71.94
78.18
83.50
87.86
89.69
92.68
93.88
95.76
96.48
97.08
97.97

98.29
98.54
99.00

100.00

33<>

PRINCIPLES OF BEHAVIOR

extent to which the upper limit of behavior oscillation exceeds the
reaction threshold, i.e., the value of the superthreshold effective
reaction potential ( S E R — S L R ) , is obtained by merely subtracting
10 from each of the b Er values; these values are shown in column
III of Table 10; they are indicated graphically in Figure 71 by
the extent to which the up-ended, bell-shaped oscillation distribu-
tions project above the reaction threshold. This portion of the
several distributions has been shaded to facilitate identification.
The value of 8 Er — bLr is next divided by the value of the stand-
ard deviation of the oscillation (o 0 ), i.e., by 10; the resulting ratios
are given in column IV of Table 10.

Finally, the probability of reaction evocation at each reinforce-
ment may be found by means of columns c and d of Table 9 (p.
311). For example, the c-value in Table 9 nearest to .141 is .1,
for which the d, or probability, value is .2 per cent. Similarly,
the c-value nearest to 1.119 is 1.1, which has a probability or
d-value of 7.45 per cent, and so on. These probability values are
given in column V of Table 10.

The moral of the rather complicated method of calculating p
just described is that: probability of reaction evocation is a normal
probability ( ogival ) function of the superthreshold magnitude of
effective reaction potential. If this hypothesis, coupled with the
growth hypothesis of the relation of the number of reinforcements

bHr (and so to 8 Er), yields learning curves conforming sub-
stantially to those observed under corresponding empirical con-
ditions, all hypotheses involved in the derivation will tend in so far
to be substantiated.

THE REACTION-EVOCATION LEARNING CURVES IMPLIED BY THE

THRESHOLD-OSCILLATION HYPOTHESIS

A graphic representation of the progressive movement across the
reaction threshold of the zone of reaction-potential oscillations as
a whole is shown in Figure 71. In each normal distribution the
per cent that the shaded area stands to the entire area under the
bell-shaped curve (column V of Table 10) is the probability (p)
that an adequate stimulation will evoke the conditioned response.
Because of their special significance these latter values are repre-
sented by a separate graph, which is shown in Figure 72. Actually
this curve closely parallels an extensive group of empirical learn-
ing curves, a fact which strongly tends to substantiate the thresh-

REACTION THRESHOLD-RESPONSE EVOCATION 331

old-oscillation hypothesis. A good example of this type of empirical
learning curve from a simple conditioning situation is that of
Hilgard and Marquis, shown in Figure 73. An example from a
much more complicated experimental situation is given above, in
Figure 68 (p. 307).

It is highly probable, however, that many learning curves of
this general character reported in the literature have their shape
determined in part at least by other factors. Presumptive exam-
ples of sigmoid learning curves so complicated are seen in Figure 24

Fig. 72. Graph showing a theoretical probability-of-reaction-evocation
curve of learning plotted as a function of the number of reinforcements. Note
the somewhat distorted sigmoid shape of the learning curve as contrasted
with the simple growth function shown in Figure 71. The extent of the
distortion is indicated by the unequal separation of the three vertical lines
drawn through the curve.

(p. 108). One of these complicating factors is the quasi-normal
distribution in the learning difficulty of the various elements which
are learned, e.g., the several syllables of rote series, the learning
scores of which are pooled in the plotting of learning curves. An-
other source of the initial phase of positive acceleration sometimes
found in learning curves is the interaction of the oscillation of two
competing reaction potentials observed under certain conditions of
simple trial-and-error learning (Figure 24 and p. 107 ff.).

It is quite evident that the characteristics of the curves of
both Figure 72 and Figure 73 are radically different from those

PRINCIPLES OF BEHAVIOR

332

of the simple positive growth function to which certain empirical
learning curves approximately conform (see Figures 21 and 23,
p. 103 ff.) and which we have found reason to believe also parallels
the functional relationship of habit strength to the number of rein-
forcements. The relationships of the two contrasting types of
learning curve may perhaps best be understood by comparing Fig-
ure 72 with the simple positive growth function from which it was
derived; this is shown as the upper bounding curve in Figure 71.
At a glance the probability-of-reaction-evocation curve is seen to
be approximately sigmoid or ogival in form. A closer examination

Fio. 73. An empirical probability-of-evocation type of learning curve
showing a roughly sigmoid shape. This graph represents the frequency of
conditioned lid reactions in dogs on each successive day of conditioning*
(Adapted from Hilgard and Marquis, 4 , p. 112.)

shows, however, that this learning curve differs from the true ogive
(Figure 70, p. 313) in a number of important respects. In the first
place, the probability-of-reaction-evocation function ( p ) has an
initial section which is horizontal, standing at zero probability for
three successive stimulations. This phenomenon has already been
formulated as Corollary I. An empirical parallel to it may be
observed in the initial portions of the curve shown in Figure 73.

A second and analogous difference lies in the fact that the
probability-of-reaction-evocation curve has also a final phase of
indefinite extent in which it is quite horizontal at 100 per cent, even
though learning in the sense of increase in habit strength and

REACTION THRESHOLD-RESPONSE EVOCATION 333

reaction potential progresses steadily (Table 10 and Figure 71).
These considerations lead to the formulation of a second corollary:

II. In the original simple conditioning or learning of an all-or -
none type of reaction , the maximal level of 100 per cent of reaction
evocation may occur in the later stages of reinforcement even
though the reaction potential may steadily increase through con-
tinued reinforcement.

It may be noted in connection with Corollaries I and II that
the probability of reaction evocation (p) ceases to be an indicator
of reaction potentiality both when the zone of oscillation is
wholly below the reaction threshold and when it is wholly above
it (Figure 71). For this reason, probability of reaction evocation
in these extreme ranges is an entirely inadequate indicator either
of habit strength or of effective reaction potential. Where reactions
of the all-or-none variety are being learned, there are available,
however, two supplementary measures throughout the considerable
upper range of reaction potential which yields a uniform 100 per
cent of reaction evocations. These are: (1) reaction latency (9),
and (2) resistance to experimental extinction (11).

A further examination of Table 10 and Figure 72 shows that
the probability-of-reaction-evocation learning curve differs from
a true ogive (Figure 70) in that the steepness of rise of the first
half is relatively greater than that of the second half. For example,
the curve of Figure 72 rises from the last zero probability (after
two reinforcements) to 50 per cent probability (after about 11 rein-
forcements) as the result of 11 — 2 or 9 reinforcements. On the
other hand, it requires about 17 reinforcements (28 — 11) to pass
from the 50 per cent level to the 100 per cent level. The true ogive
of the normal probability function is, of course, quite symmetrical
(see Figures 70 and 75). The asymmetry of the near-ogival prob-
ability-of-reaction-evocation learning curve is due to the influence
of the progressively slower rise of the learning growth function
(bEr) as it approaches its asymptote. Thus we arrive at our third
corollary:

III. The probability-of-reaction-evocation type of learning
curve , while roughly resembling the normal ogive function , clearly
deviates from it in that as the probability of reaction evocation
increases from zero to 100 per cent , the rate of rise of the prob-
ability of reaction evocation is progressively slower as compared
with the corresponding portion of the normal ogive.

It ia important to note in this connection that the initial period

334 PRINCIPLES OF BEHAVIOR

of positive acceleration of the learning curve is to be expected on
the present set of hypotheses only when the learning starts from
absolute zero, or when the rate of learning is relatively slow, or
when both these conditions obtain. Calculations have been made
of the values of p when at the outset b^r (possibly through gen-
eralization of excitation) is
supposed to stand at 7 wats
rather than at zero and when
the fractional learning incre-
ment is taken at the relatively
large value of 1/10 rather
than at 1/20; these p values
are represented graphically in
Figure 74. There it may be
seen that under the assumed
conditions a fairly conven-
tional growth-type curve of
learning is to be expected,
with scarcely any suggestion
of the initial period of positive
acceleration shown in Figure
72. It is believed that this is
the explanation of the failure
of many reactions of the all-
or-none type to show the gen-
erally ogival form of learning
curve. In this connection it
may be recalled that Thur-
stone found the sigmoid form of learning curve only when the
material to be learned was difficult (10).

The preceding bit of analysis accordingly brings us to our
fourth corollary:

IV. If the learning of an all-or-none type of reaction sets out
with an bE r value appreciably above zero, and if the rate of learn-
ing is relatively rapid , the initial period of positive acceleration
characteristic of this type of learning will not appear in the prob-
ability -of -reaction-evocation curve of learning.

As a brief summary of the foregoing examination of the rela-
tionship of the theoretical construct, effective reaction potential
(bEr), to reaction evocation as based on the threshold-oscillation
hypothesis, we present in Figure 75, in comparable graphic form,

Fio. 74. A theoretical probability-of-
reaction type of learning curve resulting
from the assumption of a rapid rate of
habit acquisition together with a sub-
stantial reaction potential from general-
ization at the outset of the reinforce-
ments. See text for details.

REACTION THRESHOLD-RESPONSE EVOCATION 33 5

the two functions which combine to produce the sigmoid theoretical
curve of learning shown in Figure 72. The upper portion of Fig-
ure 75 represents g Eit as the familiar simple positive growth func-

Fiq. 75. Graph showing the analysis of the theoretical learning curve of
Figure 72 into two components connected by the mediating construct b E b .
The upper curve shows bEb as a growth function of N, and the lower curve
shows p as an ogival function of bEb - bL m .

tion of the number of reinforcements (N). The lower portion of
the figure shows the critical second component, the probability of
reaction evocation, as an ogival function of JS Rt i.e., of the effective
reaction potential less the reaction threshold. This latter type of
function is the special concern of the present chapter, since it
serves to anchor the construct g T2 R to an observable consequent
event.

33 *

PRINCIPLES OF BEHAVIOR

REACTION LATENCY (a£*) AS A FUNCTION OF EFFECTIVE

REACTION POTENTIAL ( s ^ r )

Proceeding with our systematic examination of the relationship
of effective reaction potential to quantitatively observable response

Fia. 76. Graphic representation of the theoretical components of one of
Simley’s empirical curves of learning (Figure 22) which is plotted in tr<rms of
reaction latency (ats). The upper component is the familiar curve of habit
strength (and so of bEb) plotted as a simple growth function of the number
of reinforcements (N). The lower component represents ats as a function of
b Er. The broken line extending upward toward infinity as bEb values grow
less than about 24 wats represents an extrapolation into the region below the
reaction threshold which in this case was attained after a single reinforcement.

phenomena, we consider next the functional relationship of a&R
to reaction latency, sf*, the time intervening between the begin-

REACTION THRESHOLD-RESPONSE EVOCATION 337

ning of the stimulus and the beginning of the response. Unfortu-
nately, the quantitative aspects of this function are complicated by
the conditions of reinforcement in a manner not yet fully deter-
mined. It is well known, for example, that given suitable condi-
tions of learning, organisms can be trained to react after a consid-
erable range of predetermined delays. However, if the promptness
of the reinforcement is dependent upon the promptness of the reac-
tion, there is in general an inverse functional relationship of the
number of reinforcements to the reaction latency. The following
analysis assumes these latter learning conditions.

This relationship is perhaps best represented by means of a
graphic analysis of the Simley ( 9 ) reaction-latency learning curve
shown in Figure 22. Such an analysis is presented in Figure 76
in a manner parallel to that of the purely theoretical probability-
of-evocation function shown in Figure 75. Thus in the upper por-
tion of Figure 76 there appears the familiar positive growth function
representing the relationship of the effective reaction potential to
the number of reinforcements. In the lower portion of the figure
we note the functional relationship which is of present interest,
that of g t R to g E R . A glance at this portion of the figure shows
that at R is a negatively accelerated decreasing function of 8 E R
where bEr is greater than about 24 wats, the empirical reaction
threshold.

RESISTANCE TO EXPERIMENTAL EXTINCTION (n) AS A FUNCTION

OF EFFECTIVE REACTION POTENTIAL (s^r)

Continuing our systematic examination of the relationship of
effective reaction potential to quantitatively observable response
phenomena, we attempt as our third task the determination of the
functional relationship of gE R to the number of unreinforced reac-
tion evocations (n) required to extinguish a reaction potential to
a given degree of impotence, say to three successive stimulations
which fail to evoke observable reaction. Here, much as in the case
of reaction latency, the quantitative aspects of the problem are
complicated by the conditions of reinforcement. It has been shown
by Humphreys (7), for example, that the course of extinction is
rather different when the reinforcements of the original learning
have been accompanied by a considerable number of non-reinforced
stimulations. The present analysis is accordingly somewhat ten-
tative, as was that concerned with B ts- It proceeds on the assump-

338 PRINCIPLES OF BEHAVIOR

tion that the conditions of reinforcement are constant throughout
and that they are uncomplicated by non-reinforcement or other
disturbing factors.

This relationship may perhaps best be illustrated by means of

Fia. 77. Graphic representation of the theoretical components of William^
empirical learning curve (Figure 23) which is plotted in terms of the number
of unreinforced reaction evocations (n) required to produce experimental
extinction. The upper component is the usual curve of habit strength (and
so of s Eb ) plotted as a simple growth function of the number of reinforce-
ments ( N ). The lower component represents n as a simple linear function
of aEs*

a graphic analysis of Williams’ learning curve shown in Figure 23.
This curve, it will be recalled, represents the number of unrein-
forced reactions (n) required to produce a given degree of extinc-
tion as a function of the number of reinforcements, drive remaining

REACTION THRESHOLD-RESPONSE EVOCATION 339

constant. The theoretical analysis of this empirical function into
two components in parallel with the analysis presented in Figures
75 and 76, is shown in Figure 77. As in Figures 75 and 76, the
upper portion of this figure represents reaction potential as the
familiar positive growth function of the number of reinforcements.
The lower curve shows as the second component that the number
of extinction reactions (n) is a simple increasing linear function
of the reaction potential (sE R ).

It may be noted incidentally that, according to this function, n
becomes negative when g E R has values less than about 8 wats. This,
of course, is an expression of the reaction threshold emphasized in
connection with the interpretation of Figure 50.

INTENSITY OR AMPLITUDE OF REACTION (i) AS A FUNCTION OF

EFFECTIVE REACTION POTENTIAL ( s ^ r )

As our fourth and concluding analysis of the relationship of
effective reaction potential to quantitatively observable response
phenomena, we shall consider that of B E R to a variety of reaction
which is not of the all-or-none type. We have chosen for this pur-
pose Hovland’s empirical curve of the acquisition of a conditioned
galvanic skin reaction, shown as Figure 21 (p. 103). In connection
with this analysis it must be pointed out, much as in the cases of
reaction latency and of extinction, that the function may possibly
be complicated by variations in the conditions of the original
acquisition of the reaction potential. It is known (p. 305 ff.) that
striated-muscle reaction may easily be trained to a particular am-
plitude by special conditions of reinforcement. However, since
no evidence, experimental or observational, has been found of such
a tendency in the case of either the galvanic skin reaction or
salivary secretion, distorting complications of the functional rela-
tionship of A to b Er, where the autonomic nervous system is pri-
marily involved, seem rather unlikely. At all events, it is assumed
in the following analysis that no such complications are involved.

The components into which this learning curve breaks up are
presented in Figure 78 in a manner exactly parallel to Figures 75,
76, and 77. The upper portion of Figure 78 shows, as usual, the
effective reaction potential ( B E R ) as a simple positive growth func-
tion of the number of reinforcements (AT). The lower portion of
the figure, our chief concern here, reveals that the amplitude of the
conditioned galvanic skin reaction is a simple linear increasing

340

PRINCIPLES OF BEHAVIOR

function of the effective reaction 'potential. It thus resembles very
closely the relationship of g E R to the number of unreinforced reac-
tion evocations (n) required to produce extinction. However, a
striking difference is also to be noted: whereas in Figure 77 the
straight line originates below the reaction threshold, in Figure 78
it originates an appreciable distance above it; this reflects the

. Fro. 78. The analysis of a curve of the conditioning of the galvanic
skin reaction (Figure 21) into two components connected by the theoretical
construct aSi.

well-known fact that previous to specific conditioning, almost any
stimulus will evoke the galvanic skin reaction. Stated in another
way, this means that at the outset of the learning process here
under consideration (Figure 21), the reaction tendency is well
above the reaction threshold, just as that shown in Figure 50 is
appreciably below it.

REACTION THRESHOLD-RESPONSE EVOCATION

34 1

THE COMPETITION OF SIMULTANEOUS INCOMPATIBLE REACTION

POTENTIALS

There remains to be considered one more important matter
before the relationship between b Er and reaction is formally com-
plete. It will be recalled that when discussing the reaction thresh-
old we pointed out above that no reaction will be evoked unless
bEr is greater than sL«. It must now be noted that reaction will
not inevitably occur when B E R exceeds bLr, or even when the mo-
mentary effective reaction potential ( s 2?r) is greater than b Lr.
This is because there are frequently encountered situations in which
the stimulus complex impinging on the organism may simultane-
ously give rise to two or more incompatible reaction potentials.
Examples of such reaction tendencies would be opening and clos-
ing the eyelids, extending and flexing the arm or leg, or speaking
almost any two words of a language. It is obvious that in such
a situation the momentarily weaker of two competing reaction
potentials cannot possibly mediate its reaction.

Whether in such cases the reaction potential of the dominant
tendency is completely brought to bear in the evocation of the
reaction, or whether it suffers some diminution resembling that
long known as associative inhibition, is not known. Certain ob-
servations suggest that the latter supposition is the true one. At
all events, the experimental evidence has led some investigators
(3, p. 206) to the view that the interference of incompatible reac-
tion potentials may be mutual and that this generates an inhibitory
potential which behaves in many, if not all, respects as does that
generated by experimental extinction. Unfortunately the dynamics
of this very common situation have not been sufficiently investi-
gated to warrant an attempt at a detailed quantitative statement,
particularly as to possible indirect inhibitory effects.

Ignoring for the present, then, possible generalized inhibitory
tendencies which may result from the competition of incompatible
reaction potentials, we state as a first approximation that if two or
more superthreshold reaction potentials exist in an organism at the
same instant, only the reaction of that one whose oscillation value
at the moment is greatest will be evoked. Thus is formally con-
cluded the task of anchoring the construct b Er to objectively ob-
servable behavior on the consequent side.

34 2

PRINCIPLES OF BEHAVIOR

SUMMARY

The pivotal theoretical construct of the present system is that
of the effective reaction potential ( a l? R ). An attempt has been
made in the preceding chapters to anchor this in a secure and quan-
titative manner to antecedent observable conditions of habit for-
mation, of motivation, and of stimulation immediately preceding
reaction evocation. The task of the present chapter has been to
anchor it in a parallel manner on the posterior or consequent side.
The observable phenomena available for this purpose are found in
four aspects of reaction evocation: (1) the probability (p) of reac-
tion evocation; (2) reaction latency { a t R ) ; (3) resistance to experi-
mental extinction (n) ; and (4) in the case of autonomically medi-
ated reactions, reaction amplitude (A). This means that a success-
ful performance of our task involves the quantitative determination
of the functional relationship of B 7? R to p, a t R , n, and A, respec-
tively.

Of the four relationships, the one at present offering the best
prospects of a successful conclusion is that involving the probability
of reaction evocation, p. This is in part because the various addi-
tional constructs necessarily involved are revealed in a fairly obvi-
ous manner by relatively independent considerations. The con-
structs in question are (1) the reaction threshold ( a L R ) and (2) the
downward oscillation (sO«) to which effective reaction potential
( a E R ) is believed to be subject.

So long as the maximum reaction potential lies below the reac-
tion threshold, no activity can be evoked. However, as this maxi-
mum passes the reaction threshold in the more simple learning
situations, the probability of reaction evocation increases pro-
gressively until the zone of oscillation is wholly above the reaction
threshold, when the probability of reaction will be 100 per cent,
or perfect. Since it may require several reinforcements to raise
b E r to a value exceeding that of a L R , it often happens that there is
an initial region of uniformly zero reaction probability in learning
curves. Similarly, the zone of maximum oscillatory interference
with a E R (minimum value of B E R ) may have passed above the
reaction threshold before the limit of habit acquisition afforded by
the conditions of reinforcement is reached; this may bring about
a more or less protracted terminal period of uniformly perfect
response probability (100 per cent) in this type of learning curve.
Finally, owing to the presumably normal distribution of the oscil-

REACTION THRESHOLD-RESPONSE EVOCATION 343

lation function, its passage across the reaction threshold yields
during the learning process a probability-of-reaction-evocation
learning curve which possesses when complete a distinctly skewed
ogival form. The general agreement of learning curves secured
under corresponding empirical conditions, with the three theoreti-
cal implications just listed, gives considerable additional substan-
tiation to the belief in the general soundness of the various hy-
potheses involved.

One of the results emerging from the above analysis is a further
confirmation of the critical hypothesis that B H R (and so B E Rf in
case D is constant) is a simple positive growth function of the
number of reinforcements ( N ). This last consideration is of stra-
tegic importance because it enables us to determine in an indirect
manner the presumptive functional relationship of each of the three
remaining response phenomena to the standard or maximal value
of effective reaction potential ($2^*). It happens that the sample
empirical learning curve based on each of the three response aspects
is expressible to a reasonably close approximation by equations,
all of which contain an explicit positive growth function of N. The
replacement of this expression by its equivalent, 8 E R , leaves the
particular response phenomenon stated in terms of B E R . This type
of analysis when applied to the learning curves based on presum-
ably representative sets of empirical data indicates as a first ap-
proximation that: reaction latency ( a t R ) is a negatively accelerated
decreasing function of effective reaction potential ( 8 Er) ; that
resistance to experimental extinction (n) is an increasing linear
function of effective reaction potential ( B E R ) ; and that amplitude
of reaction (mediated by the autonomic nervous system) is an in-
creasing linear function of effective reaction potential ( B E R ).

However, situations frequently occur where the stimulus com-
plex impinging on the organism at a given moment is such as to
give rise to two or more incompatible reaction potentials. Ignoring
for the present the possibility of indirect inhibitory effects, we
may say as a first approximation that in such situations the reac-
tion potential which is strongest will dominate the others by evok-
ing its reaction. Thus is formally completed the anchoring of the
constructs B E R and B E B to objectively observable phenomena on
the consequent side.

Generalizing on the considerations elaborated above, we now
formulate six additional primary principles:

344

PRINCIPLES OF BEHAVIOR

POSTULATE 11

The momentary effective reaction potential (s&r) must exceed the
reaction threshold ( sLr ) before a stimulus (S) will evoke a given re-
action ( R ).

POSTULATE 12

Other things equal, the probability (p ) of striated-muscle reaction evo-
cation is a normal probability (ogival) function of the extent to which
the effective reaction potential ( sEr ) exceeds the reaction threshold
( sLr ).

POSTULATE 13

Other things equal, the latency ( stR ) of a stimulus evoking a striated-
muscle reaction is a negatively accelerated decreasing monotonic function

of the momentary effective reaction potential ( sEr ), provided the latter
exceeds the reaction threshold ( sLr ).

POSTULATE 14

Other things equal, the mean number of unreinforced striated-muscle
reaction evocations (n) required to produce experimental extinction is a
simple linear increasing function of the effective reaction potential ( sEr )
provided the latter at the outset exceeds the reaction threshold (sLr)»

POSTULATE 16

Other things equal, the amplitude (A) of responses mediated by the
autonomic nervous system is a simple linear increasing function of the

momentary effective reaction potential (sEr).

POSTULATE 16

When the reaction potentials ( sEr ) to two or more incompatible re-
actions (i?) occur in an organism at the same time, only the reaction

whose momentary effective reaction potential ( s Er ) is greatest will be
evoked.

Postulate 16 completes our statement of primary principles.

NOTES

Mathematical Statement of Postulate 11

(Six) 'XeS • sEr > sLr’ID '( 3y) -yeR

( 46 )

REACTION THRESHOLD-RESPONSE EVOCATION 345

Mathematical Statement of Postulate 12

’ = 0 when bEr ^ sLr _

sLr — sEr + 2.5 a

= m 0 5) - <

)]

^ = 1 when s&r ^ + 5 <t,

where <r is the standard deviation of sOr,
and

^(a) = f <M0 eft,

— CD

where «#»(t) is the standard probability function.

Mathematical Statement of Postulate 13

(47)

&Ir =

sfl

6 ' *

where a * and b' are positive empirical constants.

(48)

Mathematical Statement of Postulate 14

n = c’sE r - S', (49)

where c ' and /' are empirical constants.

Mathematical Statement of Postulate 15

A = WrEr - i' t (50)

where W and V are empirical constants.

Mathematical Statement of Postulate 16

(3w)’WeaER l : (3x)-xtsER i : (STy)-yoS: bEr x > bEr 1 > bL r : 3 : (gz)-zeRi (51)

The Oscillation Component of the Empirical Reaction Threshold

This component would arise in a simple reinforcement determination of the
reaction threshold (as in the experiment cited from Hill, 6) in the following
manner: As the (maximum) value of bEr I s passing the reaction threshold, it is
in the highest degree improbable that rEr will be_depressed to the minimal (zero
or near-zero) degree of oscillation the first time bEr is tested for reaction failure.
For example, on the assumption that the maximum range of oscillation is 5 a,
the chances are 95 to 5 against &Er exceeding the reaction threshold even when
the standard or maximal value of rEr has risen above the reaction threshold by
as much as one-fifth its range of oscillation, i.e., nearly as much as that shown
at N = 6 in Figure 71. This means, of course, that with a limited number of
trials at this level of learning there will inevitably be an appreciable mean value
of bEr previous to the first reaction evocation. It is evident that the extent to
which this mean value of bEr at the first reaction evocation exceeds zero must be

PRINCIPLES OF BEHAVIOR

346

included in the magnitude of the empirical reaction threshold as indicated above
from conditioning data. Indeed, it would produce a considerable superficial
indication of an empirical reaction threshold even though no inertial reaction
threshold (bLr) were present.

By exactly analogous reasoning it is easy to show that a parallel effect would
result from the determination of the empirical reaction threshold by the pooling
of results secured in the ordinary extinction of all-or-none reactions.

The Equations From Which the Curves in Figure 75 Were Plotted
The equation of the upper or growth function is,

bEr = 100 (1 - 10-- 02226 *).

The equation expressing the functional dependence of p upon &Er is,

f = 0 if sEr ^ sLr __

pi = 1.0125J*(2.5) - *(3.5 - .1 bEr)\

[ = 1 if rEr ^ &Lr + 5 <r,

where a is the standard deviation of the oscillation function ( bOr )» bLr is taken
as 10 wats, and 5 a is taken as 50 wats. This is the first of the four critical equations
anchoring the construct B E R on the consequent side to objectively observable phenomena.
The second of the three right-hand members of the above expression represents
the normal ogive function.

The Equations From Which the Curves in Figure 76 Were Plotted

The equation fitted to one of Simley’s composite empirical learning curves
{Figure 22 plotted in terms of reaction latency) (afe) is,

, _ 5.11

BR = [100 (1 - io--i«*)]V

Inspection of the right-hand member of this equation reveals the expression
(100(1 — 10 -122 *) which is the familiar expression of sEr as a positive growth
function of N. This circumstance enables us directly to write the equation,

bEr = 100 (1 - 10 -- 1 * 2 *),

from which was plotted the curve in the upper portion of Figure 76.

Replacing 100 (1 — 10 -122 ^) in the first equation by bEr, we have as the
other component, '

from which was plotted the lower curve of Figure 76. This is the second of the
four critical equations anchoring the construct bEr on the consequent side to objectively
observable phenomena. It should be noted that, strictly speaking, this equation
does not hold for sEr values below about 24 wats, which in this case marks the
reaction threshold.

REACTION THRESHOLD-RESPONSE EVOCATION 347

The Equations From Which the Curves of Figure 77 Were Plotted

The equation fitted to Williams* and Perm's empirical learning curve (Figure
23) is,

n = .66 1100 (1 - 10- 018*)] - 4.

By inspection, the bracket in the right-hand member of this equation is a positive
growth function of N and presumably represents bEr. We accordingly may
write the equation,

sF* = 100 (1 - 10- 018*).

It is from this equation that the upper curve of Figure 77 is plotted.

Replacing the bracket of the original fitted equation by bEr, we may write
from equation 43,

n = (B — W)( s E r — sLr -)

where'B = 116.6, W = 42.5, c = 12.5, and L = 1,

as the remaining component of the empirical equation. It is from this equation
that the lower curve of Figure 77 has been plotted. This is the third of the four
equations anchoring the critical construct , bEr, on the consequent side to objec-
tively observable phenomena.

It should be noted in this connection that certain evidence, such as that repre-
sented by Figure 57 and Table 6, appears superficially to be in conflict with the
linear relationship of n as a function of bEr, represented by the above equation
and in Figure 77. It is possible that the apparent disagreement is due to the
fact that both Table 6 and Figure 57 are based on processes under autonomic
control, whereas Figure 77 presupposes striated muscle response. Because of
the seeming inconsistency in the evidence, Postulate 14 and the above equation
expressing it cannot be accepted without reservation until definite confirmatory
evidence becomes available.

The Equations From Which the Curves of Figure 78 Were Plotted

The equation fitted to Hovland’s empirical learning curve data (Figure 21) is,

A]= .141 [100 (1 - 10- W3*)] + 3.1.

By inspection, the bracket’may be seen to enclose a simple positive growth func-
tion of N. Accordingly we write directly the equation,

bEr = 100 (1 - 10- 033 *).

Replacing the bracket of the original equation by bEr, we have as the second
component,

A = .UIsEr -f- 3.1.

It is from this that the lower curve of Figure 87 has been plotted. This is thu
last of the four equations anchoring the critical construct bEr on the consequent sid w
to objectively observable phenomena.

348

PRINCIPLES OF BEHAVIOR

REFERENCES

1. Boring, E. G. A history of experimental psychology. New York: Cen-

tury Co., 1929. Now published by D. Appleton-Century Co., Inc.

2. Fulton, J. F. Muscular contraction and the reflex control of movement .

Baltimore: Williams and Wilkins Co., 1926.

3. Gibson, E. J. A systematic application of the concepts of generalization

and differentiation to verbal learning. Psychol. Rev^ 1940, Iff, 196-229.

4. Hiloard, E. R., and Marquis, D. G. Conditioning and learning. New

York; Appleton-Century Co., Inc., 1940.

5. Hell, C. J. Retroactive inhibition in conditioned response learning. Ph J>.

thesis, 1941, Yale University.

6. Hovland, C. I. The generalization of conditioned responses. IV. The

effects of varying amounts of reinforcement upon the degree of generali-
zation of conditioned responses. J. Exper . Psychol ., 1937, 21, 261-276.

7. Humphreys, L. G. The effect of random alternation of reinforcement on

the acquisition and extinction of conditioned eyelid reactions. J. Exper .
Psychol., 1939, 26, 141-158.

8. Perin, C. T. Behavior potentiality as a joint function of the amount of

training and the degree of hunger at the time of extinction. J. Exper.
Psychol., 1942, SO, 93-113.

9. Sim ley, O. A. The relation of sub limina l to supraliminal learning. Arch.

Psychol., 1933, No. 146.

10. Thurstone, L. L. The learning curve equation. Psychol. Monog., 1919,

26, No. 114.

11. Williams, S. B. Resistance to extinction as a function of the number of

reinforcements. J. Exper. Psychol ^ 1939, 23, 506-521.

CHAPTER XIX

The Patterning of Stimulus Compounds

In our account of the “law of reinforcement” (p. 71 ff.) it was
pointed out with care that reactions are conditioned to the central
afferent impulses (s) set in motion by the action of stimulus ener-
gies (S) upon receptors. Because of the approximate one-to-one
relationship between S and s, in many instances the influence of
the principle of afferent neural interaction (p. 42 ff.) was largely
ignored in the interest of expository simplicity. For similar rea-
sons the distinction between b Hr and ,H R was not stressed, and
the habit strengths to the evocation of a given reaction associated
with different stimulus elements or aggregates occurring together
were reported in the main, though not exclusively (p. 216 ff.), as
summating by a kind of diminishing returns principle (p. 223 ff.).
Moreover, a detailed account of the role of afferent neural inter-
action, particularly in the patterning of stimulus compounds, could
not be given until the reader had been familiarized with the major
phenomena of conditioned inhibition (p. 282 ff.), of its generaliza-
tion (p. 283 ff.), and of oscillation (p. 304 ff.). Now that all of our
primary principles have been set forth, we may present in a little
detail some further important implications of the principle of
afferent neural interaction.

For reasons presently to be disclosed (p. 374) there are many
situations in which organisms would have about as good chance
of survival if they responded as if their reactions were conditioned
to stimulus elements (S) as they would if their reactions were
conditioned to central afferent impulses (s) ; moreover, habits do
in fact summate (p. 209 ff.). There are, however, innumerable
situations in which the response must be made to a certain com-
bination or configuration of circumstances ( 2, p. 503) and in which
that response would not be reinforced if given to any single cir-
cumstance as represented by isolated stimulus elements or aggre-
gates 1 or in any combination. For example, a red light suspended

1 Actually, of course, stimulus elements probably never literally occur
alone; a stimulus element or aggregate is said to occur “alone” when its
onset occurs alone, the onset of the other stimuli having occurred earlier or
later. A stimulus element is a stimulus energy which activates a single re-

349

350

PRINCIPLES OF BEHAVIOR

over a street intersection will cause a man to halt when his goal
would lead him to cross the street, but a red light in a drugstore
window will not cause him even to slow his pace; he responds not
to the red light alone, but to it as a component in a particular
combination of stimulus aggregates. Now, as a rule, learning to
react, or not to react, to a stimulus combination as distinguished
from its components is more difficult than the simple conditioning
of a reaction to a stimulus compound. This learning to respond to
stimulus combinations or configurations, as such, we shall call the
patterning of the stimulus compound in question.

For purposes of convenience we shall distinguish two main forms
of stimulus patterning. The one to be considered first, because con-
ceptually the simpler, is that in which the onset of the several
stimulus energies involved takes place at the same time; this is
called simultaneous stimulus patterning. The second form is that
in which the onset of the several stimulus energies occurs succes-
sively; this is called temporal stimulus patterning. In some cases
the response will be reinforced when it follows the presentation of
the stimulus combination; this results in positive patterning. In
other cases, reinforcement will occur only when response follows
the separate presentation of the stimulus components; this results
in an extinction of the tendency for the compound to evoke the
reaction and is therefore called negative patterning.

SOME EXPERIMENTAL EXAMPLES OP SIMULTANEOUS STIMULUS

PATTERNING

The first experimental attack on the problems of the pattern-
ing of stimulus compounds began in the laboratory of Pavlov,
though he did not use this expression in connection with the experi-
ments. Unfortunately he gives no example of simultaneous pat-
terning. He does, however, make this summarizing statement:

It was noticed that if a conditioned reflex to a compound stimulus
was established . . it was easy to maintain it in full strength and at
the same time to convert its individual components, which gave a positive
effect when tested singly, into negative or inhibitory stimuli. This re-
sult is obtained by constant reinforcement of the compound stimulus,
while its components, on the frequent occasions when they are applied
singly, remain without reinforcement. This experiment can be made with

ceptor organ. A stimulus aggregate may be defined as a group of stimulus
elements all of which usually begin and terminate at the same time, e.g., the
stimulus energies given off by an object such a a an apple.

THE PATTERNING OF STIMULUS COMPOUNDS

351

equal success in the reverse direction, making the stimulatory compound
into a negative or inhibitory stimulus, while its components applied singly
maintain their positive effect. ( 8 , p. 144.)

Fortunately there have recently become available some detailed
examples of closely analogous forms of simultaneous patterning in
dogs; these are from a study reported by Woodbury { 10 ). In one
experiment a dog named “Dick” was placed in a wooden stock
much like that employed by Pavlov. Just in front of the dog’s
head was a light horizontal wooden bar; when this bar was raised
about half an inch it closed an electrical circuit, activating an
electromagnetic device which released into a pan before the dog
a pellet of appetizing food. The animal was first taught to lift the
bar with his nose to secure the food. Later a wooden shutter was
placed before the bar to prevent the dog from nosing it except
immediately after the stimuli to be patterned were given. These
stimuli were produced by two buzzers, one with a low-pitched,
rather raucous tone (represented by L), and the other with a defi-
nitely higher-pitched and much less raucous tone (represented by

TABLE 11

Quantitative Statement or the Course of the Patterning of a Simul-
taneous Stimulus Compound Learned by Woodbury's Dog, “Dick," To-
gether With the Steps in the Derivation of the Patterning Coefficient.
(Derived from an unpublished table from Woodbury’s data, 10.)

Successive

Hundreds

Differential

Reinforce-

ments

Per Cent
Reaction
Evocation
by 50
Presen-
tations of
Compound

(Q)

100

Per Cent
Reaction
Evocation
by 25
Presen-
tations of
Component
H

Per Cent
Reaction
Evocation
by 25
Presen-
tations of
Component

100

Mean

Per Cent
Reaction
Evocation
by H and L

Patterning

Coefficient,

Q-Q

(Q)

{P)

91.5

-.5

100.0

0.0

100.0

0.0

100.0

0.0

88.0

12.0

50.0

30.0

70.0

20.0

80.0

20.0

80.0

12.5

87.5

100.0

2.0

98.0

100.0

PRINCIPLES OF BEHAVIOR

35 1 2

H). The apparatus was so set that when the buzzers were sounded
simultaneously ( HL ) and the dog nosed the bar, the food would
be delivered, but when either buzzer was sounded alone (H or L)
and the dog nosed the bar, no food would be delivered.

According to Pavlov’s summary statement quoted above, this
differential reinforcement, as it is called, should differentiate the
compound from the components. Table 11 and Figure 79 show
that such a differentiation did in fact occur and that positive pat-
terning resulted. The dog learned practically never to nose the
bar when either H or L was presented alone but practically always

Fig. 79. Graph showing in detail the course of the learning of Woodbury’s
dog, “Dick,” to react positively to a simultaneous stimulus compound con-
sisting of a high-pitched buzzer (//) and a low-pitched buzzer (L), and not to
react to the components, i.e., the buzzers presented separately. Of each one
hundred trials, 50 were of the compound ( HL ), 25 were of the high-pitched
component, and 25 were of the low-pitched component. This figure was
plotted from the values shown in Table 11. (Adapted from Woodbury, 10.)

to nose it at the presentation of the compound, HL; this process,
however, was a protracted one requiring some 1300 trials. An
examination of Figure 79 reveals, in addition to the fact of final
differentiation or successful patterning, the following details of the
learning process:

1. Although before the introduction of differential training the animal
was responding 100 per cent of the time to the stimulus of the bar fol-
lowing the lifting of the shutter, soon after this training began both the
compound and the components showed some tendency to extinction or
generalized extinction effects, as indicated by the initial depression of all
these curves.

2. As practice continued, all three curves rose to 100 per cent, where
they remained for some 300 trials.

THE PATTERNING OF STIMULUS COMPOUNDS

353

3. Following this recovery, each component stimulus gradually lost its
power to evoke the reaction, the high-pitched buzzer distinctly less slowly
than the other.

4. The general shape of the falling curves approximates roughly that
of the ogive; i.e., at first the fall is positively accelerated, after which the
acceleration reverses itself and becomes negative, finally approaching the
horizontal at a zero frequency of response.

In full verification of Pavlov’s assertion quoted above, Wood-
bury reports the detailed record of the pattern learning of a second
dog, “Chuck,” in which the component stimulus elements in the
experimental arrangement just described were positive and the com-

Fiq. 80. Graph showing in detail the course of the learning of Woodbury's
dog, “Chuck,” to react positively to the components of a simultaneous audi-
tory stimulus compound ( H ) and ( L ) and to react negatively to the com-
pound itself ( HL ). Out of each 100 presentations, 50 were HL, 25 were H,
and 25 were L.

pound was negative. The course of this bit of negative pattern
learning is shown in some detail by the curves of Figure 80. A
comparison of this figure with Figure 79 reveals the following:

1. After the first shock of the non-reinforcement associated with the
differential training, the positive components rose to 100 per cent and
maintained that position substantially to the end of the training, much
as did the positive compound in Figure 79.

2. Upon the whole, the components (positive) showed more of a
tendency to fall below 100 per cent in Figure 80 than did the compound
(positive) in Figure 79.

3. While the negative compound of Figure 80 began to lose its effective
excitatory potential more promptly than the negative components in
Figure 79, the former showed a greater resistance to complete extinction
than the latter.

354

PRINCIPLES OF BEHAVIOR

PRINCIPLES UPON WHICH THE PATTERNING OF STIMULUS

COMPOUNDS IS BASED

While Pavlov was completely familiar with the various aspects
of the experimental phenomena of patterning, it would appear that
he regarded patterning as an ultimate molar phenomenon in that
he did not succeed in breaking it down into more elementary prin-
ciples. This is a little surprising, since he was well acquainted
with most, if not all, of the principles required for doing so, notably
the principle of afferent neural interaction (see Chapter III, p.
42). Because of the importance of patterning for behavior dy-
namics, such a derivation will presently be given.

The reader is now familiar with the molar principles upon
which the theoretical derivation of the patterning of stimulus com-
pounds is based. For convenient reference they are summarized
as follows:

o. Habits are connections between receptor discharges and effector
discharges, as shown by the following diagram (p. Ill),

S > s -» r » R,

in which the arrow with the broken shaft between s and r represents
the habit, or H of the symbol B H R (Postulate 4, p. 178 ff.).

6. Afferent impulses (s) interact, changing all impulses involved to a
greater or less degree, while on their way to that point in the nervous
system at which the reinforcement process makes the junction between
the s and the r; the expression 5 represents the s as changed by inter-
action with another s (Postulate 2, p. 47 ff.).

c. When s has become s, this represents a change in position on a
generalization continuum (p. 188). Thus s would tend to evoke a reaction
conditioned to s, though because of the generalization gradient the habit
strength, sH R , and so the reaction potential aE R , would be weaker than
gE R (Postulate 5).

d. In the case of complete experimental extinction, the total inhibition
is equal to the difference between the reaction potential and the reaction
threshold, 8 L R (p. 323 ff.), i.e.,

Ir = B&R — fiAs*

e. The gradients of generalization of both excitatory potential, b Er>
and inhibitory potential, a I Rt have approximately the same slope;
their equations have the same exponent (p. 275 ff.).

/. Other things equal, the magnitude of the interaction effects between
afferent impulses from a single receptor field, e.g., the retina, will be
greater than that between impulses which arise from distinct receptor
fields, e.g., the retina and the tactile receptors.

THE PATTERNING OF STIMULUS COMPOUNDS

355

g. Other things equal, the magnitude of the interaction effect of one
afferent impulse upon a second is an increasing monotonic function of the
magnitude of the first (Postulate 2, p. 47 ff.).

h. When a stimulus energy ceases to act on a receptor, the condition-
able afferent impulses set in motion by the stimulation continue to be ac-
tive in the central neural substance for some seconds; the intensity of
this perseverative tendency gradually diminishes as a negative growth
function of the time since the termination of the action of the stimulus
on the receptor (Postulate 1, p. 47).

t. The empirical degree of stimulus patterning (P) where the reaction
is of the all-or-none type (such as that displayed by Woodbury’s dogs)
may be represented sufficiently well for our present expository pur-
poses by the formula,

P = Q~Q, (52)

where Q represents the mean empirical per cent of reaction evocations by

the reinforced phase of the compound, and Q represents the mean per cent
of reaction evocations by the negative or non-reinforced phase of the
compound.

The theoretical degree of stimulus patterning (P') may be con-
veniently represented for our present purposes by the formula,

P” = 100 (l - (S3)

in which Q' is the effective reaction potential of the negative portion of
the discrimination, and Q' is the effective reaction potential of the posi-
tive portion of the discrimination.

Both the P and the P ’ formulas yield a value of 100 for perfect
or complete patterning, and one of zero where no patterning ten-
dency exists. In case patterning is negative, both formulas yield
a negative value. Several examples of the use of the second, or P'
formula, will be encountered in the immediately following pages.
An extended example of the use of the empirical or P formula is
contained in Table 11. Substituting appropriately in this formula
from the first row of entries, we have,

P = 91 - 91.5
= - . 5 ,

which yields a slight and presumably atypical negative value; this
value constitutes the first entry in the last column. The pattern-
ing index or coefficient, P, reveals effectively the course of the
acquisition of patterning as a unified process. This is brought
out strikingly by the graphic representation of the P-values as a

3 5*

SUCCESSIVE HUNDREDS OF DIFFERENTIAL REINFORCEMENTS

Fia. 81. Graphic representation of the progress of the positive patterning
of a simultaneous stimulus compound learned by Woodbury’s dog, “Dick.”
Plotted from the final column of Table 11. Note the approximately ogival
form of this curve. (See Chapter XIII, p. 204 ff.)

function of the number of differential reinforcements, which may
be seen in Figure 81.

TH BORE TTCAIj derivation of the spontaneous or quasi-

patterning OF SIMULTANEOUS STIMULUS COMPOUNDS

Proceeding to the application of the above principles to specific
values, we shall first make a theoretical analysis of the tendency
to spontaneous or differentially unlearned patterning exhibited by
a simple conditioned simultaneous stimulus compound in which the
components are potentially negative, i.e., about to be unreinforced.
In this way will be rounded out a discussion begun above (Chap-
ter XIII, p. 204) concerning the influence of afferent interaction
on the dynamics of stimulus compounds (p. 217 ff.).

Let it be assumed that by the Pavlovian technique a simultane-
ous stimulus compound of two components (S t and S,) is condi-
tioned to a reaction (R) strongly enough for S t to command a

THE PATTERNING OF STIMULUS COMPOUNDS

357

habit strength of 40 habs and S$ to command a habit strength of
60 habs. Taking the value of the effective drive, D (Chapter XIV,
p. 245), at unity we have,

and

"ifia = 40 wats

JjEr =60 wats.

Now, by the physiological summation of these two excitatory poten-
tials, we have,

+ 40 + 60 -^60

= 100-24
.*. Q* = 76 wats.

We next consider the excitatory potential of the two compo-
nents if each is acting alone, i.e., independently. Since the respec-
tive afferent impulses will not be interacting upon the separate
presentations of S x and £*, the afferent impulses resulting from their
individual action on the receptors will be represented simply by
s x and s g , i.e., without the breve. Now (Postulate 5), the shift from
the Si of the compound to the s x of separate action will involve a
generalization effect and therefore a fall in excitatory potential from
i.E R to , E r . Let it be assumed that throughout this particular
problem, unless otherwise stated, the reduction from s to s, or vice
versa, amounts to 25 per cent. Accordingly, we have as the sepa-
rate action of the respective components,

tJ E B — 40 — 10 = 30 wats

t2 E B = 60 — 15 = 45 wats;

and, by calculated physiological summation, these two potentially
negative stimulus components would exert the equivalent of,

.! + at E R = 30 + 45 -

30 X 45
100

= 75 - 13.5
Q' = 61.5 wats.

358

PRINCIPLES OF BEHAVIOR

Next, substituting in the formula for calculating the theoretical
patterning index (P')> we have,

p. ,

= 100 (1 - .809)

= 100 X .191
= 19.1.

This means that at the very outset of the patterning process , in
case the components are negative , afferent interaction produces a
small amount of spontaneous positive patterning .

We turn next to the theoretical analysis of the spontaneous
patterning of a simultaneous stimulus compound in which the com-
ponents are separately reinforced and the compound is potentially
negative. In that event, making assumptions analogous to those
of the preceding case, we have,

and

hE r = 40 wats,
sjEr = 60 wats,

which if summated physiologically would yield,

Q ~ n + = 76.

. . . f process of generalization from s to s, each ex-

citatory potential in passing from the separate state to that of
the compound would reduce the reaction potentials from 40 to 30

wats, and from 60 to 45 wats respectively, which by physiological
summation would yield,

. . , _ O' = + - 61.5,

just as before.

Substituting these Q-values in the formula for patterning, we
have,

P' = 19.1,

exactly as when the compound was reinforced. These considera-
tions lead us to our first corollary:

I. Following the application of positive reinforcement hut pre-
vious to the application of unreinforcement, situations involving
simultaneous stimulus compounds will show a certain tendency to

THE PATTERNING OF STIMULUS COMPOUNDS

359

patterning, the strength of the tendency being the same whether
the compound or the components are reinforced.

We turn now to the consideration of the influence on the amount
of spontaneous patterning of increasing the amount of neural inter-
action effect. Clearly, the greater the amount of interaction, the
greater will be the reduction in generalization from s to s and vice
versa. Let it be supposed that this reduction is increased from
the 25 per cent assumed above, to 50 per cent. In that case, on
assumptions otherwise the same as above, if the compound is posi-
tive we have after generalization,

= 40 - 20 = 20,

By physiological summation, assuming the components
tive, we have,

Q = 20 + 30
= 50-6
= 44.

20 X 30
100 ■)

to be nega-

Since, exactly as when the generalization reduction was 25 per cent,

Q = + = 76,

we have, by substituting in the patterning formula,

P , = 100 (i _ g)

" = 100 (1 - .578)

= + 42.2.

This value of -f- 42.2, when the generalization reduction due to
neural interaction is 50 per cent, is to be compared with the smaller
P- value of 19.1 when the generalization reduction is 25 per cent.
Since, by Corollary I, the spontaneous tendency to patterning is the
same when the components are positive, it will not be necessary to
derive that case for the decreased rate of generalization.

Generalizing from the above calculation, we arrive at our sec-
ond corollary :

II. The greater the amount of afferent neural interaction, the
greater unit be the amount of “spontaneous” patterning, both posi-
tive and negative.

360

PRINCIPLES OF BEHAVIOR

THEORETICAL DERIVATION OF THE POSITIVE PATTERNING OF

SIMULTANEOUS STIMULUS COMPOUNDS BY
DIFFERENTIAL REINFORCEMENT

The theoretical derivation of quasi or spontaneous patterning
has much in common with that of the genuine patterning of stim-
ulus compounds which is attained by means of differential rein-
forcement. Accordingly the above account of the former will serve
as a useful introduction to the slightly more complicated derivation
of the latter now to be given. The two processes have in common
a dependence upon afferent interaction with the consequent opera-
tion of the generalization gradient, physiological summation, etc.
Genuine patterning involves the additional complication of the gen-
eration of experimental extinction ( I R ) which necessarily results
from the non-reinforcement that is a part of differential reinforce-
ment. 1 There is also involved the associated generation of condi-
tioned inhibition ( sI R ) and the generalization of a portion of this

back upon the positive or reinforced phase of the patterning situa-
tion.

We shall take as our first example the case in which the com-
pound is positive (reinforced). In order to make the exposition
as intelligible as possible, we shall assume, as in the above exam-
ples, that we have a compound of only two stimulus components,
Si and S g , and that the process begins with the initial conditioning
of the compound to the reaction, R; this yields, as before, excitatory
potentials as follows:

'i 1 E B = 40 wats

'i t E R = 60 wats,

which, by physiological summation, yields a 'i t +'i g E R of 76 wats
(p. 223). Now, by generalization, % g E B shrinks 25 per cent from
a value of 40 to one of 30 and • t E B shrinks from a value of 60 to
one of 45, i.e.,

and

m x E b — 30 wats
t3 E B — 45 wats.

Assuming a reaction threshold of 10 wats as a minimum neces-
sary to evoke reaction (see p. 323 ff.), the non-reinforcement of the

1 The generation of inhibition during reinforcement is here neglected in
the interest of expository simplicity.

THE PATTERNING OF STIMULUS COMPOUNDS 361

separate components tl E R and 8g E R to the reaction threshold (10
wats) would reduce each to a value of zero responses, with the
generation of inhibition amounting to,

30 — 10 = 20 pavs

and

45 — 10 = 35 pavs.

Assuming that half of each of these values is made up of condi-
tioned inhibition (sI R ) and therefore is subject to stimulus gen-
eralization (p. 281 ff.), and recalling that, by assumption, 75 per
cent stimulus generalization occurs between s and s, or vice versa,
we have as the amount of inhibitory potential generalizing from
the respective non-reinforcements back upon the corresponding
reinforcement phase,

X .75 = 7.5 pavs
X .75 = 13.13 pavs.

Summating these inhibitory potentials physiologically, we have,

= 7.5 + 13.13

7.5 X 13.13
100

= 20.63 - .99
= 19.64 pavs.

From this it follows from the formula for B E R that,

'irrtJBa = 76 - 19.64
= 56.36,

Q' = 56.36 wats.

On the other hand, since the two threshold values left by the extinc-
tion process never operate together, they will be averaged, rather
than summated, i.e.,

sy _ 10 + 10
W " 2

= 10 wats.

3& 2 PRINCIPLES OF BEHAVIOR

Substituting these Q-values in the formula for patterning, we have,

* - 100 (* - sis)

= 100 (1 - .177)

= 82.3,

which represents a large degree of patterning.

Suppose, on the other hand, that we employ the empirically
more realistic response formula for observed patterning,

~ P - Q - Q.

The per cent of responses to 9 jE r equals 10 wats (after extinction
to zero), and » t E s equals the same. But,

sLr — 10 ,

and when

sEr = sLr,

V = 0 ,

i.e.,

Q = 0.

Next, if we assume that the maximum range of oscillation ( gOn )
is 40, since,

Q' = 56 wats

and the threshold is taken as 10 wats, it follows that oscillation

will frequently carry the reaction potential below the reaction
threshold ( B L R ) , i.e.,

56 - 10 > 40,

from which it follows that the will vield 100 per cent reac-

tion evocations, and the value of Q will be 100 per cent. Substi-
tuting these values in the formula for P, we have,

P = Q-Q

= 100 - 0.00
= 100 ;

i.e., empirical patterning (P) will be perfect. Thus we arrive at
our third corollary:

THE PATTERNING OF STIMULUS COMPOUNDS 363

III. In case a simultaneous stimulus compound and its com-
ponents receive differential reinforcement, the components being
negative ( unreinforced ), sufficient training will produce complete
empirical patterning provided the differences between s and s are
great enough to bring about a residual which exceeds the

range of behavioral oscillation ( a O R ) plus the reaction threshold

{sI/r)’

THEORETICAL DERIVATION OF THE NEGATIVE PATTERNING OF

SIMULTANEOUS STIMULUS COMPOUNDS BY
DIFFERENTIAL REINFORCEMENT

We turn next to the case of the patterning of a simultaneous
stimulus compound in which all the conditions are assumed to be
the same as those in the preceding example, except that the com-
pound is negative, i.e., St and S s are reinforced separately until

n E R = 40 wats

and

l2 E R = 60 wats.

Then S t and S t are presented as a compound and given differential
reinforcement. By reasoning exactly analogous to that of the pre-
ceding example, s will generalize to the s of the compound to the
extent of 75 per cent, yielding, when in the compound at the out-
set of differential reinforcement,

'i l E R = 30 wats

and

'iJEn = 45 wats.

Since these are extinguished jointly in each will contribute

roughly equal amounts to the 10-wat threshold, i.e., Q f = 10. This
will leave to be extinguished by the negative aspect of differential
reinforcement, and so to be converted into inhibition,

30 — 5 = 25 pavs

and

45 — 5 = 40 pavs.

364

PRINCIPLES OF BEHAVIOR

If, now, half of this inhibition in each case is bIr and 75 per
cent of J R generalizes, we have,

n 1 * =

X .75 = 9.38 pavs,
X .75 = 15.00 pavs

to generalize back on 8j E r and 80 E R respectively. It follows that,

n Em = 40 - 9.38 = 30.62 wats

and

^Er = 60 — 15 = 45.00 wats.

Since the two negative potentials are generated together, they are
already eliminated physiologically at the reaction threshold, i.e.,

Q' = 10 wats.

On the other hand, since 8i E r and 8b E r never operate together, they
will be averaged:

G , = 30.62 + 45.00
H 2

•\ Q' = 37.81 wats.

Substituting these Q-values in the patterning formula, we have,

p ' - 100 (' - sra)

= 100 (1 - .264)

= 100 (.736)

= 73.6,

which represents a considerable patterning tendency.

Suppose, now, we consider the response aspects of this situa-
tion. It may be seen that, assuming the range of behavioral oscil-
lation (sOn) to be 40 wats, since, as stated above,

sE R = 45 wats,

and since

sL r = 10 wats,

and

45 - 10 < 40,

THE PATTERNING OF STIMULUS COMPOUNDS 365

S a will not evoke R 100 per cent of its presentations but, by Table
9, would yield about 96 per cent of reaction evocations. 1 Also
since

'IjEr = 30.62 wats,

and since

s L r = 10 wats,
32.62 - 10 < 40,

it follows that St will evoke R at less than 100 per cent of its
presentations, i.e., at approximately 61 per cent of the stimulations.
Thus,

96 + 61
2

78.5.

Accordingly, substituting in the empirical patterning
have,

P = 78.5 - 0.00

formula we

= 78.5,

a degree of response patterning appreciably less than perfect and
so definitely less than the degree of response patterning found
under comparable conditions where the compound was positive. It
is evident, however, that if BjHr an d were sufficiently rein-

forced to bring each of them up to a value of about 70 habs, both
9 j E r and ag E R would mediate the evocation of R on 100 per cent
of the trials, in which case the empirical patterning index would
be 100, i.e., perfect even under the conditions of component rein-
forcement. This brings us to the statement of our fourth corollary:

IV. In case a simultaneous stimulus compound and its compo-
nents receive differential reinf or cement , the compound being unrein-
forced, sufficient training will produce complete empirical patterning
provided the differences between s and s are great enough to produce
a residual B E R which exceeds the range of behavioral oscillation
{ B 0 R ) ; but, other things equal, this type of learning will require a
greater number of differential reinforcements to attain a given de-
gree of patterning than will that in which the components are unre-
inforced.

1 The statement of the method of determining this value and the reasons
for so doing were indicated above, p. 328 ff. It may be said here, however,
that the major assumption underlying the procedure is that the value of
bEb oscillates below its expressed potential magnitude and that the calcu-
lation involved the use of the probability function of behavioral oscillation
shown in Table 9 (p. 311).

3 66

PRINCIPLES OF BEHAVIOR

The tendency of empirical patterning to break down through
the failure of the components to evoke the reaction at 100 per cent
of the stimulations is nicely illustrated in Figure 80, where, at the
twelfth hundred differential reinforcements, we have,

At the same time,

96 + 100
2

98 per cent.

Q = 8 per cent,

which yields as an index of patterning,

P = 98 - 8 = 90,

a value distinctly less than perfection.

THE POSITIVE PATTERNING OF SIMULTANEOUS STIMULUS

COMPOUNDS UNDER CONDITIONS OF
DECREASED GENERALIZATION

We now come to the final step in our analysis of the process
of stimulus-patterning learning by differential reinforcement. Let
us consider the influence on this process of an increase in the affer-
ent interaction effects sufficient to steepen the fall in the general-
ization gradient from 25 per cent to 50 per cent. In order to sim-
plify the exposition we shall take the original case of a simultaneous

stimulus compound in which the compound is first reinforced so
that,

*i Er = 40 wats

and

'1 7 Er — 60 wats.

Now, by generalization, with a loss of 50 per cent, it follows that,

and

$ypR = 20 wats

= 30 wats. '

The separate extinction of each of these to the reaction threshold
of 10 wats would generate inhibition in the amount of

20 — 10 = 10 pavs

30 — 10 = 20 pavs.

and

THE PATTERNING OF STIMULUS COMPOUNDS 367

Assuming that half of each of these generalizes to the extent of
50 per cent, we have,

'zJk = ™ X .50 = 2.5 pavs

and

— on

s/s = ~2 x .50 = 5.0 pavs.

Summating these inhibition values,

CJ + V* = 25 + 5 0 “

2.5 X 5.0
100

It follows that

= 7.5 - .01
= 7.49 pavs.

i 1 + J = 76 - 7.49
= 68.51,

i.e.,

Q' = 68.51 wats.

As in the example involving 25 per cent generalization, the thresh-
old reaction potentials remaining in $ t E R and $ M E R are 10 wats
each.

Q'=^ =

10 wats.

Substituting in the following formula, we have,

p ' = 100 C 1 - SOi)

= 100 (1 - .146)

= 100 X .854

= 85.4.

This value, it will be noted, is larger than the 82.3 yielded by the
smaller fall in the generalization gradient.

However, just as in the case of the 25 per cent generalization
reduction, both components, despite the threshold values of 10 pavs,
will yield 0.00 per cent of reaction evocation, i.e.,

0 00 ± = 0 . 00 .

368

PRINCIPLES OF BEHAVIOR

On the other hand, since

68.51 - 10 > 40,

the compound will evoke R on 100 per cent of the presentations,
i.e.,

Accordingly,

Q = 100.

P = 100 - 0.00

= 100 ;

i.e., empirical patterning will be perfect. This brings us to our
fifth corollary:

V. In the patterning of stimulus compounds , the greater the jail
in the generalization gradient between 5 and s, the less will be the
difficulty in attaining a given degree of discrimination.

SOME EXPERIMENTAL EXAMPLES OF THE LEARNING OF

TEMPORAL STIMULUS PATTERNS

We turn now to the problem of the patterning of temporal
sequences of stimuli, i.e., the learning of temporal stimulus pat-
terns. Pavlov, in whose laboratory this type of learning seems first
to have been studied, describes several such experiments. In an
experiment performed by Dr. Eurman, a dog was presented with
a sequence of three stimuli: a light (L), a cutaneous stimulus (C),
and the sound of bubbling water (S). When these stimuli were
given in the order LCS , the presentation was always followed by
food reinforcement, but when they were given in the reverse order,
i.e., SCL, the combination was never reinforced. The dog finally
reached a point of training such that it secreted an average of
about 8 drops of saliva to LCS, and zero drops to SCL , which pre-
sumably indicated perfect patterning. Unfortunately, Pavlov does
not report the course of the learning of this or any other pattern,
though he makes a general statement that such learning often re-
quires protracted training and when achieved is very unstable ( 8 ,
p. 147).

Fortunately we have in the study by Woodbury {10) already
referred to, a detailed report of the course of both the positive
and the negative forms of temporal patterning. The same appa-
ratus arrangement and general procedure were employed as de-
scribed above for the patterning of simultaneous conditioned com-

THE PATTERNING OF STIMULUS COMPOUNDS 369

pounds, but with this difference: the high-pitched buzzer was
sounded for one second; then, after a pause of one second, the
low-pitched buzzer was sounded for one second, following which
the shutter was raised, giving the dog access to the bar. If the
dog nosed the bar, the act was reinforced by food. On the negative
side, the high-pitched buzzer w’as presented for one second twice
in succession, with a pause of one second between presentations.
On other occasions the low-pitched buzzer would be presented twice
in the same way. After each of these presentations the shutter
was lifted, but if the dog nosed the bar no food would be given.
The learning behavior of the dog “Ted,” by this procedure, is shown
in detail by Figure 82. A study of this figure indicates in general

GROUPS OF 100 TRIALS

Fio. 82. Graph showing in detail the course of the learning of Woodbury s
dog, “Ted,” to react positively to a temporal stimulus compound of a high-
pitched buzzer followed by a low-pitched buzzer (H,L), and negatively to a
parallel presentation of the high-pitched buzzer and of the low-

pitched buzzer (.LJj). Out of each 100 presentations, 50 were if A 25 were
HJi, and 25 were LJj. (Adapted from Woodbury, 10.)

a striking agreement with the course of the patterning of simul-
taneous stimulus compounds, except that the amount of training
required to complete the process of temporal patterning was notice-
ably greater. There is the same initial depression of all three
curves at the outset of differential reinforcement, and the same
subsequent rise of all three to 100 per cent; following this the two
non-reinforced component combinations gradually fall toward zero,
taking a roughly ogival course.

The exact reverse of the experiment just described was carried
out by Woodbury with the dog “Bengt”; i.e., the temporal se-
quences H,H and L,L were reinforced, but the sequence H,L was
never reinforced. The course of this learning may be seen in
Figure 83, a study of which reveals the same general features as
those of Figure 80, though the difficulty of learning is somewhat

370

PRINCIPLES OF BEHAVIOR

greater. It may also be noted that Figures 80 and 83 both show
a greater amount of disturbance (weakening) of the reinforced
phases than do Figures 79 and 82; this increased disturbance pre-
sumably comes from the generalization of the greater amount of
inhibition arising from the extinction of the non-reinforced com-
pound upon the relatively weak component.

It is to be noted that even though the values represented in
the preceding figures are derived from the pooling of a very large
number of observations, the performance of a single animal is not
sufficient basis for the establishment of empirical laws, though it

Fia. 83. Graph showing in detail the course of learning of Woodbury’s
dog, “Bengt,” to react positively to the temporal combinations HJI and LJj,
and to react negatively to the combination H ,L. Out of each 100 presenta-
tions, 100 were HJj, 25 were H&, and 25 were LJL. (Adapted from
Woodbury, 10.)

may serve to illustrate theoretical principles; it is as such that
Woodbury’s graphs are offered here for consideration. However,
the performances of these single animals give complete and suffi-
cient proof of one thing: they demonstrate that the positive and

the negative forms of both types of stimulus patterning can be
learned by dogs.

A THEORETICAL ANALYSIS OF TEMPORAL STIMULUS PATTERNING

The theoretical analysis of temporal stimulus patterning is at
bottom about the same as that of simultaneous stimulus pattern-
ing. This is to say that temporal stimulus patterning depends
upon substantially the same principles: afferent neural interaction,
the generalization of excitation, the extinction of the generalized
excitation, and the final generalization of this inhibition back upon
the positive reaction potential. There is, however, this important
difference: in temporal stimulus patterning the neural interaction

37 *

THE PATTERNING OF STIMULUS COMPOUNDS

presumably takes place between the afferent impulses arising di-
rectly from the second stimulation and the perseverative stimulus
traces (Postulate 1) which were originally set in motion by the
earlier stimulation. Thus the neural interaction of the components
of temporal stimulus patterns is a simultaneous affair exactly as
is that of the components of simultaneous patterning; the difference
lies in the fact of the temporal asynchronism of the action of the
respective stimulus energies which originally set the interacting
impulses in motion. From the point of view of adaptation and
survival this difference is of enormous importance, but the pattern-
ing mechanism itself differs little except quantitatively.

It is evident from the foregoing that having postulated per-
severative stimulus traces (Postulate 1), the patterning of temporal
stimulus compounds follows at once, by reasoning exactly analogous
to that by which Corollary III was derived. This brings us to the

statement of our sixth corollary:

VI. In case (1) a temporal stimulus compound and ( 2 ) a repeti-
tion of either component in the same tempo as that of the presenta-
tions of the compound , receive differential reinforcement, the repeti-
tion of the component being unreinforced, sufficient training will
produce complete stimulus patterning provided the difference be-
tween s and s is great enough to bring about a residual •
which exceeds the reaction threshold by an amount greater than the

range of behavioral oscillation {sOr)»

Having derived the basic phenomenon of temporal stimulus
patterning, we proceed to the examination of certain quantitative
differences which may, theoretically, be expected to manifest them-
selves in the comparison of the patterning of simultaneous and
successive stimulus compounds. The most striking of these dif-
ferences is evidently due to the progressive diminution in the inten-
sity of the stimulus trace with the passage of time, as stated in
Principle h (Postulate 1) ; i.e., other things equal, the intensity of
a stimulus trace will be weaker than the original afferent impulse
set in motion by the action of the stimulus energy (S) on the
receptor. Also, by Principle g (Postulate 1), a weak afferent im-
pulse (s t ) will produce a smaller interaction effect on a second
afferent impulse (s t ) than will a strong value of s x . But the
smaller the difference between s and s, the less will be the fall in
the generalization gradient; and, by Corollary V, the smaller the
fall in the generalization gradient between s and s, the greater will
be the difficulty in the attainment of a given degree of pattern dis-

37 2 PRINCIPLES OF BEHAVIOR

crimination. From these considerations there follow our seventh
and eighth corollaries:

VII. Other things equal, a given degree of the patterning of
temporal stimulus compounds will require more differential rein-
forcements than will that of simultaneous stimulus compounds.

VIII. Other things equal, in the patterning of a temporal stim-
ulus compound, the greater the lapse of time between the termina-
tion of one stimulus and the beginning of the next, the greater
will be the number of differential reinforcements required to attain
a given degree of patterning.

THE RESOLUTION OF HUMPHREY’S ARPEGGIO PARADOX

A good deal of misunderstanding has arisen in learning theory
regarding the role performed by the stimulus, presumably because
of the hidden nature of neural interaction effects. This may be
illustrated by an experiment reported by Humphrey ( 5 , pp. 198,
237):

Subjects were trained ... to raise their hand at the sound of a certain
■pecific tone. The training was accomplished by administering an electric
shock when the tone was sounded, but never when any other tone was
sounded. Suppose that the active tone was G above middle C. We have
then a conditioned reflex to this tone, which has been differentiated out so
that no other tone producible on the apparatus was followed by response.
By prolonged practice this conditioned reflex became very highly stabil-
ized. Suppose now that the active note is included in a melody or an
arpeggio such as that formed by the successive notes C, E, G, C, where
G is the active note. [The instrument was of the xylophone type with
metal cylinders, which were not damped after a note was struck, conse-
quently the vibrations from one note persisted during the striking of the
next note, as in legato piano playing, (p. 198.)] The arpeggio then con-
tains the stimulus for the conditioned response. The records show that
while the isolated note was consistently followed by response, the same
note, repeated immediately in an arpeggio, was consistently not followed
by response. The melody “Home Sweet Home” when played in the key
of C contains the note G fourteen times. Experiments showed that sub-
jects trained to this note consistently respond to it when presented in
isolation and do not respond to it when presented in the melody. This
result is very striking in view of the fact of the fourteen repetitions of the
active note included in the melody as played, (p. 237.)

How shall these experimental observations be interpreted? The
above account seems to show that when the note G was struck,
the preceding notes C and E were also still vibrating, which would

THE PATTERNING OF STIMULUS COMPOUNDS 373

produce a simultaneous stimulus compound. In the case of the
melody, “Home Sweet Home,” presumably sometimes there would
be a simultaneous stimulus compound of the “active” note and
the two or three preceding notes combined with a temporal com-
pound made up of the impulses from these active stimuli and the
perseverative stimulus traces arising from the stimulations of the
more remotely preceding notes. Since, according to the preceding
analysis, afferent impulses and perseverative traces operate in much
the same way, an analysis of the simpler arpeggio situation will
suffice for both.

According to the neural interaction hypothesis, the afferent
impulse, s c , would be changed to the impulse s c when conjoined
with the impulses arising from Sc and S B . This alone might easily
weaken the reaction potential b q Er sufficiently to produce external
inhibition. However, it must be recalled that, by differential rein-
forcement, there had presumably been developed a considerable
amount of conditioned inhibition, b c Ir and b b Ir- Let it be sup-
posed, for example, that b 0 Er has a strength of 50 wats and that
B( Jr and g J R have strengths of 30 pavs each. Because of the pro-
gressive damping by the air, the intensity of Sc and of So will bo
reduced; this should be especially true of Sc, since it was struck
first. Therefore, both bJr and b Jr will be weakened appreciably,
bJr to 20 pavs, say, and s b Ir to 25 pavs. Now, as the result of
afferent interaction and the consequent fall in the generalization
gradient, both the excitatory and the inhibitory tendencies alike
will suffer a certain reduction in strength when they enter the com-
pound, say 20 per cent. This leaves us with the following values:

'SgEjt = 40 wats
-.Jr = 16 pavs
'SrIr = 20 pavs.

Summating the two inhibitions, we have,

16 X 20

'ic+'S* 1 * =16 + 20 fob

= 36 — 3.2

= 32.8.

374

PRINCIPLES OF BEHAVIOR

It therefore follows that the effective reaction potentiality to the
lifting of the hand at the striking of the note G in the midst of
the arpeggio will be,

sJ? R = 40 - 32.8
= 7.2.

If we assume, as in the preceding computations (p. 364), that the
reaction threshold has a value of 10 wats, it appears that under
the given conditions the net effective reaction potential available
for reaction evocation in the arpeggio (7.2 wats) would be below
the reaction threshold and therefore the hand would not be lifted
during the playing of the arpeggio, exactly as Humphrey found.
Thus Humphrey’s auditory configurational problem finds a natural
and consistent explanation in terms of habit dynamics. 1

SOME GENERAL CONSIDERATIONS CONCERNING THE FUNCTIONAL

DYNAMICS OF STIMULUS PATTERNS

From the point of view of causation, an organism and its entire
environment must be regarded as a complex causal interacting
unit (I). Probably in all adaptive situations the act of the organ-
ism does not yield an effect which produces a particular type and
amount of reinforcement until every one of a number of different
conditions is satisfied. In an ideal adaptive situation the organism
would have (a) receptors which would respond differentially to the
impact of energies characteristic of each of the several critical
conditions, and (6) receptors responsive to each of the various
critical conditions which would prevent the act from resulting in
reinforcement; also (c) the stimulus energies associated with these
several conditions should actually impinge on the relevant receptor,
e.g., they should not be shielded from the receptor by the inter-
position of some other object. Under these ideal conditions, pat-
terning would seem to be the only effective form of habit structure.
Unfortunately, even though for the most part conditions a and b
obtain with higher organisms, condition c is often satisfied not only
imperfectly but to varying and fortuitous degrees of imperfection.

1 It may be added that the context of the quotation from Humphrey
cited above seems to indicate that no small portion of the confusion in this
case comes from the purely semantic difficulty of having failed to recog-
nize, and symbolize, the distinction between the st im ulus energy (5) and the
afferent impulse (s).

375

THE PATTERNING OF STIMULUS COMPOUNDS

For this reason the question of whether a reaction in a given situa-
tion will be followed by reinforcement is almost always more or
less of a gamble for the organism, even in the most advanced stages
of training. It thus comes about, as Brunswik (/) has pointed out,
that as the number of critical stimulus cues increase, the proba-
bility becomes greater that the total group or confipration of
causal factors necessary for the act to eventuate in reinforcement
is present in fact. Therefore, as noted in connection with the habit
dynamics of stimulus compounds, the summation of habit strengths
which are based on the afferent impulses of the stimulus compo-
nents, s, as contrasted with the interaction effects represented by
s, is a primitive biological first’- approximation to a calculus of adap-
tive probability. This mechanism, coupled with that of the reac-
tion threshold ( S L«), prevents the organism from wasting its energy
by reacting when the probability of need reduction is too slight.
Similarly, if the number of times that a given stimulus element or
aggregate has been associated with reinforcement is small, the
habit strength will be small, the probability of response evocation
will be small, and so here again the organism will tend to react
automatically to the probabilities of the situation. The implica-
tions of the very numerous permutations of these and related
factors cannot be entered into here, though many of them are fairly

obvious. .. ,

But in case a situation is sufficiently stabilized for the critical

variable factors to satisfy condition c, as was the case with Wood-
bury’s dogs, experiments show, on the basis of recopiized prin-
ciples, that the organism can largely transcend the initial crude
summation calculus of adaptive probabilities by reacting, or not,
with precision to particular combinations or configurations of stim-
ulus aggregates. It is true that a reaction associated with a par-
ticular pattern discrimination developed in one static setting may
not be followed by reinforcement in another, so that the organism
must continue to gamble to some extent as long as life lasts. . How-
ever, each new reaction brings with it increased training in dis-
crimination, which makes the odds fall progressively more in the

animal’s favor as life goes on.

One favoring subsidiary factor is that by compound trial and
error organisms learn to adjust their receptors in such a way as
to expose them more adequately to all the environmental energies
relevant to a given need; this is the behavioral mechanism which
brings about searching or exploration. Another favoring behavioral

PRINCIPLES OF BEHAVIOR

37 ^

mechanism, at least in human organisms, is the conditioning by
social trial and error of characteristic symbolic acts such as words,
to certain, stable and significant stimulus aggregates (objects) ; this
presumably facilitates very greatly the indirect generalization (p.
191) of instrumentally adaptive reactions set up in one configura-
tional situation which enables them to function in other situations
adaptively similar but differing to a considerable degree in stimulus
configurational characteristics. Probably this explains why Razran
( 9 ) found with verbally sophisticated human subjects that con-
figurational generalization was considerably wider than was simple
stimulus generalization. Feeble-minded individuals and dogs might
be expected to show rather different results.

On the basis of the above considerations the conjecture is haz-
arded that one of the more important capacities in the higher levels
of intelligence is that of discriminating afferent interaction effects.
It is believed that in human beings this will be found intimately
connected with the transfer by indirect generalization of the reac-
tions from one stimulus pattern to another through the mediation
of words. At very high levels of adaptive efficiency it is expected
that words will constitute the mediating stimuli of the stimulus
patterns themselves.

SUMMARY

Many life situations require for optimal chance of survival
that the organism shall react to certain combinations of conditions
(stimulus compounds) differently than to the component conditions,
either when these components are encountered “separately” or in
other combinations. The most radical, and at the same time the
most simple, formulation of this problem is presented by Pavlov’s
experimental arrangement for the discrimination of a stimulus
compound from its components. Experiments have fully demon-
strated that organisms over a wide phylogenetic range are able to
learn such discriminations, though usually with comparative diffi-
culty.

This type of learning by organisms turns out upon analysis
definitely to be a derived or secondary phenomenon, dependent
upon a number of logically prior principles all of which have been
recognized by Pavlov. Among the more important of these are the
connection of the reaction (12) to the afferent impulse (s) set in
motion by the stimulus (5), rather than directly to the stimulus;

THE PATTERNING OF STIMULUS COMPOUNDS 377

the mutual interaction of afferent impulses; and the downward
slope of the gradient of generalization both for excitation and for
inhibition. For the derivation of temporal stimulus patterning
there is required, in addition, the principle of the perseverative
stimulus trace.

When stripped of quantitative details, the basic logic of stimulus
patterning is rather simple. The afferent impulses produced by
the components of a dynamic stimulus compound are to some extent
different when the component is acting “alone,” i.e., in a relatively
static combination, than when it is acting with the remainder of
the dynamic compound. If a reaction is conditioned to the com-
pound, the reaction potential of a given component, because of the
generalization gradient, is less when it is acting separately than
when in the compound. During the differential reinforcement
which produces this kind of learning, the generalized excitatory
potential of the components is extinguished, developing inhibition
in proportion to the reaction strength of each component. This in-
hibitory potential generalizes back upon the compound but, again,
with a reduction due to the generalization gradient. The resulting
net loss to the reaction potential at the command of the stimulus
compound is much less than its original reaction potential ; this
ordinarily leaves the stimulus compound an amount of effective
reaction potential which is well above the reaction threshold. This
difference is the basis of the discrimination, i.e., of successful pat-
terning. While the details of positive and negative simultaneous
stimulus patterning and of positive and negative temporal stimulus
patterning differ slightly, they all conform in substance to the sum-
mary account just given. The process of genuine stimulus pattern-
ing thus turns out to be at bottom the learning to discriminate
afferent interaction effects.

By an analogous use of the same set of postulates it is possible
also to deduce the phenomenon not only of genuine stimulus pat-
terning, but also of quasi or spontaneous stimulus patterning, and,
other things equal, the amount of both kinds of patterning as an
increasing function of the degree of the afferent interaction effects
mutually generated by the components. Other deductions are to
the effect that positive patterning will be easier to learn than
negative, that simultaneous compounds will be easier to pattern
than temporal, and that the longer the time interval separating
the components in a temporal stimulus compound, the more difficult
will be the patterning. The same postulates afford a rather detailed

PRINCIPLES OF BEHAVIOR

378

explanation of why a musical note to which a reaction has been
conditioned fails to evoke the reaction when given in the midst of
an arpeggio, the remaining notes of which have been made negative
by differential reinforcement. All of these deductions are in sub-
stantial agreement with such empirical observations as are at pres-
ent available.

If the organism could be certain of the occurrence of primary
receptor discharges corresponding to every relevant condition in
its environment which contributes to the determination of whether
a given response will be followed by reinforcement, all responses
in high-grade organisms might ultimately be made only to pat-
terned stimulus compounds. But since under life conditions many
elements which determine reinforcement do not activate any recep-
tor, the basis for complete and exact patterning is frequently lack-
ing; the organism must accordingly gamble on the outcome, often
with its very life at stake. However, the processes of biological
evolution seem to have produced a fairly satisfactory non-pattemed
arrangement for meeting this contingency.

Analysis suggests that the magnitude of ordinary neural inter-
action effects is such as to produce a fall of less than 50 per cent
in the generalization gradient. Under these conditions the afferent
impulse arising from a stimulus aggregate will tend to evoke the
same reaction both alone and within the compound. When in the
compound, the habit strengths (or reaction potentials) commanded
by different stimulus aggregates presumably summate by a kind of
diminishing returns principle. As a consequence, and quite apart
from any patterning, the fewer the stimulus elements conditioned
to a given reaction which chance to be present in a given stimulus
compound, the smaller the reaction potentiality will be. This is
believed to be a kind of crude but automatic biological calculus of
the probability that a reaction evoked under given circumstances
will be followed by reinforcement.

On the basis of the foregoing considerations we now formulate
two important corollaries:

MAJOR COROLLARY IV

Differential reinforcement applied to simultaneous stimulus com-
pounds results in the patterning of such compounds, either positively
or negatively according to whether the compound or the components are
reinforced.

THE PATTERNING OF STIMULUS COMPOUNDS

379

MAJOR COROLLARY V

Differential reinforcement applied to temporal stimulus compounds
results in the patterning of such compounds, either positively or negatively
according to whether the compound or the components are reinforced.

NOTES

The Formula for Calculating the Empirical Degree of Patterning in the
Case of Reactions Which Are Not of the All-or-None Type

The formula for P (Principle t), while applying fairly well to reactions of the
all-or-none type, definitely does not apply to a response such as the galvanic
Bkin reaction, or the salivary reaction, whose amplitude varies with the magnitude
of the reaction potential. A distinctly unsuccessful attempt at a formula for
this latter type of reaction tried out in an earlier study (4) was,

Pi 4- Pa

P. + a ’

where P t is the amplitude of the reaction evoked by one stimulus component, and
Pj is the amplitude of that evoked by a second. Unfortunately, the determination
of a suitable formula for the calculation of the extent of empirical patterning
with this type of reaction requires a knowledge of the physiological limit of the
amplitude of its conditioned evocation; to find this would probably be a very
laborious procedure (4, p. 108 ff.).

The Patterning of Stimulus Compounds and the Configuration Psychologies

After studying the above chapter the reader may naturally ask what the
relation of the present behavioristic treatment of the configurational problem in
learning is to that put forward by the Wertheimer branch of the Gestalt school.
While much might be said on this subject, the few words possible to devote to it
in this place may help to clarify the reader’s understanding.

Gestalt Theorie asserts that configurations are not only logically primary but
that they are somehow primordial. Indeed, if current configurationism is ever
formulated as a true scientific theory, so that its primary and secondary principles
can be clearly distinguished, it is rather likely that a statement asserting the
reality and nature of configurations will be revealed as its sole primary principle
or postulate. The present work, on the other hand, undertakes to demonstrate
that the response of organisms to stimulus configurations is logically secondary,
that it is the result of a rather complex process of learning which is mediated by
the behaviorally primary processes of (1) afferent neural interaction, (2) per-
severative stimulus traces, (3) reinforcement, (4) generalization of reaction
potential, (5) experimental extinction, and (6) generalization of inhibition.

Gestalt writers frequently leave the impression that an adequate derivation
of the reaction of organisms to stimulus configurations is a prion impossible
from behavioristic or non-consciousness principles. The position of the present
work is that such a derivation is not only possible, but relatively simple and
straightforward. Moreover, the preceding pages have presented a number of
Buch deductions, thereby showing the Gestalt a pnori claims to be mistaken.

380 PRINCIPLES OF BEHAVIOR

Meanwhile it remains to be seen whether Gestalt Theorie can itself mediate com-
parable deductions. Clearly, no dispute exists as to the genuineness or the
importance of the patterning of stimulus compounds; the difference of opinion
concerns, rather, the logical question of whether stimulus patterning is a primary
or a secondary principle. There are, of course, other differences between Gestalt
Theorie and the present approach, but they do not particularly concern us here.

It is hoped that the derivation of the major phenomena of stimulus-pattern
learning from objective, non-consciousness principles as demonstrated above,
will contribute to the dissipation of current misunderstandings among psychol-
ogists, since these are a source of such deep and painful confusion to the scientific
public. However, optimism in this connection is seriously dampened by the
conviction that the differences involved arise largely from a conflict of cultures
( 0 , pp. 18, 685 ; 7 , p. 30) which, unfortunately, are extra-scientific and are not
ordinarily resolvable by either logical or empirical procedures.

REFERENCES

1. Brunswik, E. Organismic achievement and environmental probability.

Psychol. Rev., 1943, 60, 255-272.

2. Hull, C. L. A functional interpretation of the conditioned reflex. Psy-

chol. Rev., 1929, 36, 498-511.

3. Hull, C. L. Mind, mechanism, and adaptive behavior. Psychol. Rev.,

1937, 44, 1-32.

4. Hull, C. L. Explorations in the patterning of stimuli conditioned to

the G.S.R. J. Exper. Psychol., 1940, 27, 95-110.

5. Humphrey, G. The nature of learning. New York: Harcourt, Brace and

Co., 1933.

6 . Koffka, K. Principles of Gestalt psychology. New York: Harcourt,

Brace and Co., 1935.

7. Kohler, W. Gestalt psychology. New York: Liveright Pub. Co., 1929.

8 . Pavlov, I. P. Conditioned reflexes (trans. by G. V. Anrep). London:

Oxford Univ. Press, 1927.

9. Razran, G. H. S. Studies in configural conditioning: V. Generalization

and transposition. J. Genet. Psychol., 1940, 66, 3-11.

10. Woodbury, C. B. The learning of stimulus patterns by dogs. /. Comp •

Psychol ., 1943, 36, 29-40.

CHAPTER XX

General Summary and Conclusions

In the foregoing chapters we have made a detailed examination
of much experimental evidence, and we have considered the merits
of many alternative interpretations. Such complications, while
unavoidable in a work of this kind, necessarily tend to obscure an
integrated view in which the various components of the subject
have their proper significance. No doubt the reader has sighed
more than once for the simplicity of dogmatic affirmation and for
the over-all perspective attainable by brevity. In the present
chapter we shall endeavor to make a clarifying integration of the
major conclusions scattered through the preceding exposition.

THE NATURE OF SCIENTIFIC THEORY

The major task of science is the isolation of principles which
shall be of as general validity as possible. In the methodology
whereby scientists have successfully sought this end, two proced-
ures may be distinguished — the empirical and the theoretical. The
empirical procedure consists primarily of observation, usually facili-
tated by experiment. The theoretical procedure, on the other hand,
is essentially logical in nature; through its mediation, in conjunc-
tion with the employment of the empirical procedure, the range of
validity of principles may be explored to an extent quite impossible
by the empirical procedure alone. This is notably the case in
situations where two or more supposed primary principles are pre-
sumably operative simultaneously. The logical procedure yields a
statement of the outcome to be expected if the several principles
are jointly active as formulated; by comparing deduced or theo-
retical conclusions with the observed empirical outcomes, it may be
determined whether the principles are general enough to cover the
situation in question.

Scientific theory in its ideal form consists of a hierarchy of logi-
cally deduced propositions which parallel all the observed empirical
relationships composing a science. This logical structure is derived
from a relatively small number of self-consistent primary prin-
ciples called postulates, when taken in conjunction with relevant

381

PRINCIPLES OF BEHAVIOR

382

antecedent conditions. The behavior sciences have been slower
than the physical sciences to attain this systematic status, in part
because of their inherent complexity, in part because of the action
of the oscillation principle, but also in part because of the greater
persistence of anthropomorphism.

Empirical observation, supplemented by shrewd conjecture, is
the main source of the primary principles or postulates of a science.
Such formulations, when taken in various combinations together
with relevant antecedent conditions, yield inferences or theorems,
of which some may agree with the empirical outcome of the con-
ditions in question, and some may not. Primary propositions yield-
ing logical deductions which consistently agree with the observed
empirical outcome are retained, whereas those which disagree are
rejected or modified. As the sifting of this trial-and-error process
continues, there gradually emerges a limited series of primary prin-
ciples whose joint implications are progressively more likely to
agree with relevant observations. Deductions made from these
surviving postulates, while never absolutely certain, do at length
become highly trustworthy. This is in fact the present status of
the primary principles of the major physical sciences.

BEHAVIOR THEORY AND SYMBOLIC CONSTRUCTS

Scientific theories are mainly concerned with dynamic situa-
tions, i.e., with the consequent events or conditions which, with the
passage of time, will follow from a given set of antecedent events
or conditions. The concrete activity of theorizing consists in the
manipulation of a limited set of symbols according to the rules
expressed in the postulates (together with certain additional rules
which make up the main substance of logic) in such a way as to
span the gap separating the antecedent conditions or states from
the subsequent ones. Some of the symbols represent observable
and measurable elements or aggregates of the situation, whereas
others represent presumptive intervening processes not directly sub-
ject to observation. The latter are theoretical constructs. All well-
developed sciences freely employ theoretical constructs wherever
they prove useful, sometimes even sequences or chains of them. The
scientific utility of logical constructs consists in the mediation of
valid deductions; this in turn is absolutely dependent upon every
construct, or construct chain, being securely anchored both on the
antecedent and on the consequent side to conditions or events

GENERAL SUMMARY AND CONCLUSIONS 383

Fia. 84. Diagram summarizing the major symbolic constructs (encircled
symbols) employed in the present system of behavior theory, together with
the symbols of the supporting objectively observable conditions and events.
In this diagram S represents the physical stimulus energy involved in learn-
ing; R, the organism’s reaction; s, the neural result of the stimulus; the
neural interaction arising from the impact of two or more stimulus com-
ponents; r, the efferent impulse leading to reaction; G, the occurrence of a
reinforcing state of affairs; bHr, habit strength; S, evocation stimulus on the
same stimulus continuum as 5; bHr, the generalized habit strength; Co, the
objectively observable phenomena determining the drive; D, the physiological
strength of the drive to motivate action; bEr, the reaction potential; W ,
work involved in an evoked reaction; Jr, reactive inhibition; «/», conditioned
inhibition; bEr, effective reaction potential; bOr, oscillation; bEb, momentary
effective reaction potential; bLr, reaction threshold; p, probability of reaction
evocation; latency of reaction evocation; n, number of unreinforced reac-
tions to produce experimental extinction ; and A, amplitude of reaction.
Above the symbols the lines beneath the words reinforcement, generalization,
motivation, inhibition, oscillation, and response evocation indicate roughly the
segments of the chain of symbolic constructs with which each process is
especially concerned.

which are directly observable. If possible, they should also be
measurable.

The theory of behavior seems to require the use of a number
of symbolic constructs, arranged for the most part in a single
chain. The main links of this chain are represented in Figure 84.

384 PRINCIPLES OF BEHAVIOR

In the interest of clarity, the symbolic constructs are accompanied
by the more important and relevant symbols representing the ob-
jectively anchoring conditions or events. In order that the two
types of symbols shall be easily distinguishable, circles have been
drawn around the symbolic constructs. It will be noticed that the
symbols representing observables, while scattered throughout the
sequence, are conspicuously clustered at the beginning and at the
end of the chain, where they must be in order to make validation
of the constructs possible. Frequent reference will be made to this
summarizing diagram throughout the present chapter, as it reveals
at a glance the groundwork of the present approach to the behavior
sciences.

organisms conceived as self-maintaining mechanisms

From the point of view of biological evolution, organisms are
more or less successfully self-maintaining mechanisms. In the
present context a mechanism is defined as a physical aggregate
whose behavior occurs under ascertainable conditions according to
definitely statable rules or laws. In biology, the nature of these
aggregates is such that for individuals and species to survive, cer-
tain optimal conditions must be approximated. When conditions
deviate from the optimum, equilibrium may as a rule be restored
by some sort of action on the part of the organism; such activity
is described as “adaptive.” The organs effecting the adaptive
activity of animals are for the most part glands and muscles.

In higher organisms the number, variety, and complexity of
the acts required for protracted survival is exceedingly great. The
nature of the act or action sequence necessary to bring about opti-
mal conditions in a given situation depends jointly (1) upon the
state of disequilibrium or need of the organism and (2) upon the
characteristics of the environment, external and internal. For this
reason a prerequisite of truly adaptive action is that both the con-
dition of the organism and that of all relevant portions of the
environment must somehow be brought simultaneously to bear on
the reactive organs. The first link of this necessary functional
rapport of the effector organs with organismic needs and environ-
mental conditions is constituted by receptors which convert the
biologically more important of the environmental energies (S) into
neural impulses (s). For the most part these neural impulses flow
to the brain, which acts as a kind of automatic switchboard medi-

GENERAL SUMMARY AND CONCLUSIONS 385

ating their efferent flow (r) to the effectors in such a way as to
evoke response (.R). In this connection there are two important
neural principles to be noted.

The first of these principles to be observed is that after the
stimulus ( S ) has ceased to act upon the receptor, the afferent im-
pulse (s) continues its activity for some seconds, or possibly min-
utes under certain circumstances, though with gradually decreasing
intensity. This perseverative stimulus trace is biologically im-
portant because it brings the effector organ en rapport not only
with environmental events which are occurring at the time but with
events which have occurred in the recent past, a matter frequently
critical for survival. Thus is effected a short-range temporal inte-
gration (Postulate 1, p. 47).

The second neural principle is that the receptor discharges and
their perseverative traces (s) generated on the different occasions
of the impact of a given stimulus energy (5) upon the receptor,
while usually very similar, are believed almost never to be exactly
the same. This lack of uniformity is postulated as due (1) to the
fact that many receptors are activated by stimulus energies simul-
taneously and (2) to “afferent neural interaction.” The latter
hypothesis states that the receptor discharges interact, while pass-
ing through the nervous system to the point where newly acquired
receptor- effector connections have their locus, in such a way that
each receptor discharge changes all the others to a greater or less
extent; i.e., s is changed to s t , s*, or s 3 , etc., in accordance with
the particular combination of other stimulus energies which is act-
ing on the 6ensorium at the time (see Figure 84). This type of
action is particularly important because the mediation of the
responses of organisms to distinctive combinations or patterns of
stimuli, rather than to the components of the patterns, is presum-
ably dependent upon it (Postulate 2, p. 47).

The detailed physiological principles whereby the nervous sys-
tem mediates the behavioral adaptation of the organism are as yet
far from completely known. As a result we are forced for the most
part to get along as best we can with relatively coarse molar formu-.
lations derived from conditioned-reflex and other behavior experi-
ments. From this point of view it appears that the processes oi
organic evolution have yielded two distinct but closely related means
of effective behavioral adaptation. One of these is the laying down
of unlearned receptor-effector connections ( bUb ) within the neural
tissue which will directly mediate at least approximate behavioral

PRINCIPLES OF BEHAVIOR

386

adjustments to urgent situations which are of frequent occurrence
but which require relatively simple responses (Postulate 3, p. 66).
The second means of effecting behavioral adjustment is probably
evolution’s most impressive achievement; this is the capacity of
organisms themselves to acquire automatically adaptive receptor-
effector connections. Such acquisition is learning.

LEARNING AND THE PROBLEM OF REINFORCEMENT

The substance of the elementary learning process as revealed
by much experimentation seems to be this: A condition of need
exists in a more or less complex setting of receptor discharges
initiated by the action of environmental stimulus energies. This
combination of circumstances activates numerous vaguely adaptive
reaction potentials mediated by the unlearned receptor-effector
organization ( g U B ) laid down by organic evolution. The relative
strengths of these various reaction potentials are varied from in-
stant to instant by the oscillation factor ( g O R ). The resulting
spontaneous variability of the momentary unlearned reaction poten-
tial (gO B ) produces the randomness and variability of the unlearned
behavior evoked under given conditions. In case one of these ran-
dom responses, or a sequence of them, results in the reduction of
a need dominant at the time, there follows as an indirect effect
what is known as reinforcement (G, of Figure 84). This consists
in (1) a strengthening of the particular receptor-effector connec-
tions which originally mediated the reaction and (2) a tendency
for all receptor discharges ( 5 ) occurring at about the same time
to acquire new connections with the effectors mediating the
response in question. The first effect is known as primitive trial-
and-error learning; the second is known as conditioned-reflex learn-
ing. In most adaptive situations both processes occur concur-
rently; indeed, very likely they are at bottom the same process,
differing only in the accidental circumstance that the first begins
with an appreciable strength, whereas the second sets out from zero.
As a result, when the same need again arises in this or a similar
situation, the stimuli will activate the same effectors more cer-
tainly, more promptly, and more vigorously than on the first occa-
sion. Such action, while by no means adaptively infallible, in the
long run will reduce the need more surely than would a chance
sampling of the unlearned response tendencies ($£/*) at the com-
mand of other need and stimulating situations, and more quickly

GENERAL SUMMARY AND CONCLUSIONS 387

and completely than did that particular need and stimulating
situation on the first occasion. Thus the acquisition of such
receptor-effector connections will, as a rule, make for survival;
i.e., it will be adaptive.

Careful observation and experiment reveal, particularly with
the higher organisms, large numbers of situations in which learning
occurs with no associated primary need reduction. When these
cases are carefully studied it is found that the reinforcing agent
is a situation or event involving a stimulus aggregate or compound
which has been closely and consistently associated with the need
reduction. Such a situation is called a secondary reinforcing agent,
and the strengthening of the receptor-effector connections which
results from its action is known as secondary reinforcement. This
principle is of immense importance in the behavior of the higher
species.

The organization within the nervous system brought about by
a particular reinforcement is known as a habit; since it is not
directly observable, habit has the status of a symbolic construct.
Strictly speaking, habit is a functional connection between $ and r;
it is accordingly represented by the symbol ,H r . Owing, however,
to the close functional relationship between S and s on the one
hand, and between r and R on the other, the symbol bHr will serve
for most expository purposes; the latter symbol has the advantage
that S and R both refer to conditions or events normally open to
public observation. The position of in the chain of constructs
of the present system is shown in Figure 84.

While it is difficult to determine the quantitative value of an
unobservable, various indirect considerations combine to indicate
as a first approximation that habit strength is a simple increasing
growth function of the number of reinforcements. The unit chosen
for the expression of habit strength is called the hab, a shortened
form of the word “habit”; a hab is 1 per cent of the physiological
limit of habit strength under completely optimal conditions.

CONDITIONS WHICH INFLUENCE THE MAGNITUDE OF HABIT

INCREMENT PER REINFORCEMENT

A more careful scrutiny of the conditions of reinforcement re-
veals a number which are subject to variation, and experiments
have shown that the magnitude of the habit increment (As//*)
per reinforcement is dependent in one way or another upon the

PRINCIPLES OF BEHAVIOR

388

quantitative variation of these conditions. One such factor con-
cerns the primary reinforcing agent. It has been found that,
quality remaining constant, the magnitude of the increment of
habit strength per reinforcement is a negatively accelerated in-
creasing function of the quantity of the reinforcing agent employed
per reinforcement.

A second factor of considerable importance in determining the
magnitude of A gH R is the degree of asynchronism between the
onset of the stimulus and of the response to which it is being con-
ditioned. This situation is complicated by whether or not the

stimulus terminates its action on the receptor before the response
occurs. In general the experimental evidence indicates that in
case both the stimulus and the response are of very brief duration,
the increment of habit strength per reinforcement is maximal when
the reaction (and the reinforcement) occurs a short half second
after the stimulus, and that it is a negatively accelerated decreas-
ing function of the extent to which asynchronisms in either direc-
tion depart from this optimum. In case the reaction synchronizes
with the continued action of the stimulus on the receptor, the incre-
ment of habit strength per reinforcement is a simple negative
growth function of the length of time that the stimulus has acted
on the receptor when the reaction occurs.

A third important factor in the reinforcing situation is the
length of time elapsing between the occurrence of the reaction and
of the reinforcing state of affairs ( G , Figure 84). Experiments
indicate that this “gradient of reinforcement” is a negatively accel-
erated decreasing growth function of the length of time that rein-
forcement follows the reaction. The principle of secondary rein-
forcement, combined with that of the gradient of reinforcement,
explains the extremely numerous cases of learning in which the
primary reinforcement is indefinitely remote from the act rein-
forced. A considerable mass of experimental evidence indicates
that a kind of blending of the action of these two principles gen-
erates a secondary phenomenon called the “goal gradient.” Upon
empirical investigation this turns out to be a decreasing exponen-
tial or negative growth function of the time ( t ) separating the
reaction from the primary reinforcement for delays ranging from
ten seconds to five or six minutes; delays greater than six minutes
have not yet been sufficiently explored to make possible a quan-
titative statement concerning them.

There are doubtless other conditions which influence the magm-

GENERAL SUMMARY AND CONCLUSIONS 389

tude of the increment of habit strength resulting from each rein-
forcement. Those listed above certainly are typical and probably
comprise the more important of them. An adequate statement of
the primary law or laws of learning would accordingly take the
form of an equation in which bHr would be expressed as a joint
function not only of N but of the quantity and quality of the
reinforcing agent, and of the temporal relationships of S to R and
of R to G. A formula which purports to be a first approximation
to such a general quantitative expression of the primary laws of
learning is given as equations 16 and 17, pp. 178-179.

STIMULUS GENERALIZATION

With the primary laws of learning formally disposed of, we
proceed to the consideration of certain dynamical principles accord-
ing to which habits, in conjunction with adequate stimulation ( S )
and drive ( D ), mediate overt behavior. In this connection we note
the fact that a stimulus (5, Figure 84), through its afferent im-
pulses (8, represented in Figure 84) will often evoke the reac-
tion (72) even though s may be rather different from s or I, the
receptor impulse originally conditioned to R . This means that
when a stimulus (5) and a reaction (R) are conjoined in a rein-
forcement situation, there is set up a connection not only to the
stimulus involved in the reinforcement but to a whole zone of other
potential stimuli lying on the same stimulus continuum, such as
81 , St, S 3 , and so forth. This fact, known as stimulus generaliza-
tion, is of immense adaptive significance; since stimuli are rarely
if ever exactly repeated, habits could scarcely function adaptively
without it.

Stimulus generalization has the characteristic that in general
the greater the physical deviation of S from S , the weaker will be
the habit strength which is mobilized. More precisely, the strength
of a generalized habit ( S /T*) is a linear increasing function of the
strength of the habit at the point of reinforcement and a negatively
accelerated decreasing function of the difference (d) between S
and S as measured in discrimination thresholds (j.n.d.’s). Thus
sS« is a theoretical construct anchored to the construct b Hr and
to the observables S and S (see Figure 84).

Stimulus generalization appears to take two forms — (1) quali-
tative stimulus generalization and (2) stimulus intensity general-
ization; another way of stating the same thing is to say that each

PRINCIPLES OF BEHAVIOR

390

stimulus has two generalization dimensions, quality and intensity.
Both dimensions display the negatively accelerated falling general-
ization gradient, but the qualitative gradient is markedly steeper
than is that of stimulus intensity (Postulate 5, p. 199).

PRIMARY MOTIVATION

The condition of need in organisms not only is an important
factor in habit formation, through the need reduction and rein-
forcement relationship; it also plays an important role in deter-
mining the occasions when habits shall function in the evocations
of action, the vigor of such evocations, and their persistence in the
absence of reinforcement. This is, of course, in the highest degree
adaptive, since activity such as would lead to the reduction of a
need is a sheer waste of energy when the need either does not exist
or impend.

It is a fact that the dynamic actions of many, if not all, of
the primary needs of organisms are associated with the presence
or absence in the blood of small amounts of certain potent con-
ditions or substances such as hormones. In one way or another
there are also associated with most needs certain characteristic
stimuli (So) whose intensity varies with the intensity of the need
in question (Postulate 6, p. 253).

In this connection it must be pointed out that drive ( D ) is a
logical construct, since it cannot be observed directly any more
than can effective habit strength ( a H R ). Fortunately, the more
carefully studied drives, such as hunger, thirst, and so on, have
a clear possibility of being expressed as functions of objectively
observable conditions or events; these are represented by the sym-
bol Cd in Figure 84. Since sEr is also thus anchored, the result
(gEitf Figure 84) of the combination of definite functions of these
two antecedently anchored values is itself securely anchored on
the antecedent side.

When the numerous facts of primary motivation are examined
in quantitative detail, they yield the conclusion that the poten-
tiality of response evocation ( 8 E B ) is the product of a function of
drive intensity multiplied by a function of habit strength (Postu-
late 7, p. 253). Drive stimuli (S D ) naturally become conditioned
to the reaction which is associated with reduction of the need, and
so they become an integral component of the habit ($#*) involved.
The driv* substance (or condition) in the blood appears in some

GENERAL SUMMARY AND CONCLUSIONS 391

way to sensitize the action of all habits, regardless of what drives

have been involved in their formation, in a manner which enhances

\ 9

their power of mediating reaction. The characteristic stimuli asso-
ciated with each need, through the action of stimulus patterning,
suffice to effect reactions likely to reduce whatever need is domi-
nant at the moment, rather than those appropriate to other needs
which at the time stand at or near zero.

EXTINCTION, INHIBITION, AND EFFECTIVE REACTION POTENTIAL

When a reaction is evoked repeatedly without a closely asso-
ciated drive reduction, the power of the stimulus-motivational
combination to evoke the response in question gradually dimin-
ishes; this diminution is called experimental extinction. Such a
cessation of futile activity is in the highest degree adaptive because
it tends to prevent waste of the organism’s energy reserve.

That experimental extinction does not concern merely habit
strength is shown by the fact that an increase in the drive alone
will serve to reinstate the power of stimuli to evoke a reaction
which has been extinguished to zero. A careful survey of the
numerous well-authenticated phenomena of experimental extinction
in connection with presumably related phenomena in primary moti-
vation has led to the hypothesis that experimental extinction is
to a considerable extent motivational in nature. It is supposed
that each response evocation produces in the organism a certain
increment of a fatigue-like substance or condition which constitutes
a need for rest. The mean net amount deposited at each response
appears to be a positively accelerated increasing function of work
or energy expenditure ( W ) consumed in the execution of the act

(see Figure 84). It is assumed, further, that this substance or

condition has the capacity directly to inhibit the power of S to
evoke R; for this reason it is called reactive inhibition (/*). Its
accumulation would therefore produce experimental extinction, and
its progressive removal from the tissues by the blood stream (on
the fatigue analogy) would produce spontaneous recovery.

In case the response evocations are accompanied by reinforce-
ment and occur at relatively long intervals, and ordinary motiva-
tion remains fairly high, the increase in habit strength will usually
keep the reaction potential above the reaction threshold. The
spontaneous dissipation of the inhibitory state is known to be much

392

PRINCIPLES OF BEHAVIOR

more rapid than is the ordinary loss of learning effects. This would
produce the phenomenon of reminiscence which has been especially
studied in rote learning.

Since the presence of I R constitutes a need, the cessation of
the activity which generated the need would initiate the need-
reduction process; but since need reduction is the critical element
in reinforcement, there follows with fair plausibility the molar
principle that cessation of the activity would be conditioned to any
stimuli which are consistently associated with such cessation (Pos-
tulate 9, p. 300). But a tendency to the cessation of an act would
be directly inhibitory to the performance of that act. Therefore
the inclusion of such an inhibitory stimulus in a stimulus com-
pound, the remainder of which is positively conditioned to the
response, would tend to prevent the evocation of the response in
question ; this is, in fact, the ordinary empirical test for conditioned
inhibition ( b Ir , Figure 84).

On the above view that b Ir is a negative habit, the injection of
alien stimuli into the stimulus compound would, through the prin-
ciple of afferent interaction, produce disinhibition, i.e., a temporary
reduction or total abolition of b Jr • But on the assumption that
b Ir is being set up during the process of accumulating I R , it follows
that the total inhibition (/*) at the conclusion of experimental
extinction must be in part I R and in part s /«. For this reason
disinhibition will take place only in so far as 1 R is composed of
B I R , and spontaneous recovery will take place only in so far as 1r
is composed of I R ; this means that neither disinhibition nor spon-
taneous recovery can ever restore an extinguished reaction poten-
tial to its full original strength. Other implications which flow
from the above assumptions are that there is greater economy in
distributed than in massed repetitions in rote learning, and that,
other things equal, organisms receiving the same reinforcement fol-
lowing two responses which require different energy expenditures
will, as practice continues, gradually come to choose the less labori-
ous response. This is the “law of less work.”

Implicit in the preceding discussion has been the assumption
that the reaction potential actually available for reaction evoca-
tion, i.e., the effective reaction potential ( b &b, Figure 84), is what
remains of the reaction potential ( B E R ) after the subtraction of the
total inhibition, 1 R ; i.e.,

393

GENERAL SUMMARY AND CONCLUSIONS

Since both B E B and I M are anchored to objectively observable ante-
cedent conditions, it follows that bE’m is also thus anchored.

THW OSCILLATION OF EFFECTIVE REACTION POTENTIAL

At this point it must be noted at once that, the full value of
fl TTj, is rarely brought to bear in the evocation of action. Instead,
it is subject to random or chance downward variability. These
fluctuations are believed to be due to a little-understood physio-
logical process which has the power of neutralizing reaction poten-
tials to degrees varying from moment to moment. Because of this
latter characteristic, the process is called “oscillation”; it is repre-
sented by the symbol b O r . Effective reaction potential as modified
by oscillation is called “momentary effective reaction potential ;

this is represented by the symbol B E R .

Since a O R is not directly observable, it has something of the
status of a symbolic construct; on the other hand, owing to its pre-
sumably constant value, it has less elusiveness than an ordinary
construct; it is therefore not placed in a circle in Figure 84. The
hypothetical characteristics of b O r may be listed as follows:

1. It is active at all times. # .

2. It exerts an absolute depressing action against any and all reaction

potentials, whether great or small. . A A .

3. The magnitude of this potentiality vanes from instant to instant

according to the normal probability distribution.

4. The magnitude of its action on different reaction potentials at a

given instant is uncorrelated (Postulate 10, p. 319).

Since oscillation is continuously active on all reaction poten-
tials, it plays a very great role in adaptive behavior. It pre-
sumably is responsible for many of the phenomena grouped by the
classical psychologists under the head of “attention.” It is in
large measure responsible for the fact that the social sciences must
pool many observations before ordinary empirical laws may become
manifest. Thus natural laws in the social sciences must always
be based on statistical indices of one kind or another. This in
its turn has induced much preoccupation with statistical methods
on the part of the various behavior sciences. The necessity of
pooling large numbers of observations in order to isolate empirical
laws has greatly increased the labor associated with empirical in-
vestigations and has doubtless appreciably retarded the develop-
ment of the behavior sciences.

394

PRINCIPLES OF BEHAVIOR

THE REACTION THRESHOLD AND RESPONSE EVOCATION

The anchoring on the posterior or consequent side of our chain
of behavioral constructs culminating in b Er> as shown in Figure 84,
lies in the evocation of observable reactions. In the determination
of the functional relationship of gE R to the various measurable
phenomena of responses, we encounter special difficulties owing to
the fact that B E R is itself not directly observable. If we were
quite sure of the quantitative functional relationship of B E R to its
combination of antecedent anchors, the value of B E R could be cal-
culated in empirical situations and equations could then be fitted
to the relationship of these numbers to the corresponding response
values; these equations are what we seek. Unfortunately the neces-
sary antecedent functional relationships are not yet known with
sufficient certainty.

It happens, however, that in typical sets of simple learning
results, employing the four measurable response phenemona, the
fitted equations in all cases are easily and naturally expressible
by equations involving the simple positive growth (exponential)
function of the number of reinforcements ( N ). This tends some-
what to confirm the soundness of the general growth hypothesis
of the relation of N to B H R , and so of N to 8 E R . Further inde-
pendent confirmation of the soundness of the growth hypothesis
of the relation of B E R .to N, lies in the following fact: when the
probability-of-reaction-evocation type of learning curve is ana-
lyzed theoretically, it turns out to be yielded in a degree of detail
scarcely attributable to chance on the above assumption of the
relation of 8 &r to N coupled with two additional assumptions,
each well supported by independent evidence — that of the oscil-
lation function ( a O«) and that of the reaction threshold ( 8 L R )
(see Figure 84). The characteristics of the oscillation function
have been summarized above. Moreover, the concept of the reac-
tion threshold is well established, since notions fairly comparable
to it have long been current in classical psychophysics and in
physiology. As here employed, the reaction threshold ( B L R ) is
the^ minimal amount of momentary effective reaction potential
(bEr) which is necessary to mediate reaction evocation when the
situation is uncomplicated by competing reaction potentials (Pos-
tulate 11, p. 344).

Acting, then, on the fairly well-authenticated growth hypothesis
of the relation of B E R to N , it is a relatively simple matter, by

395

GENERAL SUMMARY AND CONCLUSIONS

inspecting the equations fitted to concrete examples of the three
remaining types of learning curves and utilizing the method of
residues, to determine the functional relationship of b&r to the
particular behavior phenomena employed. As a result of this pro-
cedure it is concluded that probability of reaction evocation stands
in an ogival relationship to effective reaction potential (Postulate
12, p. 344) ; that reaction latency stands in a negatively accelerated
inverse relationship (Postulate 13, p. 344) ; and that both resist-
ance to experimental extinction and reaction amplitude (of auto-
nomically mediated responses) are increasing linear functions of

bE r (Postulates 14 and 15, p. 344).

A final complication concerning reaction evocation arises from
the fact that often the stimulus elements impinging on the receptors
at a given instant may mobilize superthreshold reaction potentials
to several different reactions, some or all of which may be mutually
incompatible. In such cases all but the strongest will necessarily
suffer associative inhibition (Postulate 16, p. 344). There are also
6ome indications that the dominant potential itself may suffer a
certain amount of blocking; indeed, this is the basis of the most
plausible theory of “forgetting” now available.

This concludes our summary of primary principles. All of these
principles are also statable in the form of quantitative equations.
This means that if the antecedent conditions S, s, R, G, t, t ' , S, $,
C D , W, 8 Or, and a L R were known, it would be possible to compute
Pt atR, n, or A by substituting appropriately in a succession of these
equations beginning on the left-hand side of Figure 84 and pro-
ceeding toward the right. For example, the calculation of g t R
would employ equations 16, 34, 44, 45, and 48.

DYNAMICS OF STIMULUS COMPOUNDS AND PATTERNS

For the most part the molar principles outlined in the preced-
ing chapters are presumably primary in nature, though occasional
secondary principles have been presented. Because of their rela-
tively primitive status in the logical hierarchy of the system and
of their especially intimate relation to survival, a few secondary
principles or mechanisms have been given special consideration
and have been listed as “major corollaries.” One of these concerns
the quantitative summation of the reaction potentials mobilized by
the several stimulus components of a stimulus compound, and the

PRINCIPLES OF BEHAVIOR

396

other concerns the matter of stimulus patterning. We shall first
take up the matter of the summation of reaction potentials.

In spite of the presumptive fact of afferent neural interaction,
the afferent discharge of each receptor contains a large amount of
similarity regardless of the influence of other stimulus elements
(and receptors) which may be active at tlje time. This means
that any stimulus component conditioned to a reaction will ordi-
narily command an appreciable potentiality to that reaction regard-
less of the other stimuli accompanying it. Now, according to the
primary law of learning, each individual receptor discharge bears
its load of habit strength, and so of reaction potential. The reac-
tion-potential loadings thus borne by the several receptor dis-
charges initiated by the different stimulus elements of a stimulus
compound presumably combine quantitatively in the same way
as do the different increments of habit strength, i.e., not by a simple
addition but according to a kind of diminishing-returns principle.
Thus if two stimulus aggregates bearing equal loads of reaction
potential to the evocation of the same response are acting simul-
taneously as a stimulus compound, their physiological summation,
quite apart from afferent interaction effects, will be less than the
arithmetical sum of the two reaction potentials; similarly, if one of
the two equally loaded stimulus aggregates making up a stimulus
compound should be withdrawn from the compound, more than
balf of the total reaction potential would remain. As a result
(except for afferent neural interaction effects), the more completely
a reinforced stimulus compound is repeated on a subsequent occa-
sion, the more likely it will be to evoke the reaction in question.

This mode of action has special adaptive significance, because
the more completely the stimulus compound is repeated, the more
similar will be the environmental situation in general to the situa-
tion in which need reduction originally occurred, and therefore the
more probably will the response in question lead again to a reduc-
tion in the need. Here we have a primitive automatic mechanism
which in effect roughly gauges the probability of a given stimulus
situation's yielding need reduction in case a given response is
evoked. This adaptive mechanism has the great advantage of
being instantly available at the presentation of any stimulus situa-
tion, novel or otherwise.

The other secondary principle mediating the response of organ-
isms to stimulus compounds, which we have included in the present
work, is that known as patterning. This operates concurrently

397

GENERAL SUMMARY AND CONCLUSIONS

with the summation principle just discussed but is much slower in
its action. However, given sufficient time for the rather difficult
learning process to take place, stimulus patterning may be very
precisely adaptive. It is a fact that in very large numbers of
situations the question of whether or not a given response will be
followed by reinforcement depends upon the presence or absence
of a particular combination of physical circumstances and so, for
the organism, upon a particular combination or pattern of stimulus
elements, rather than upon the presence or absence of any of the
components. Since each combination of stimulus elements will
modify to some extent the afferent impulses produced by each
stimulus component, any change in the stimulus compound will
also modify to some extent the afferent responses initiated by all
the remaining stimulus components. In the process of the irregular
alternation of reinforcement and extinction called differential rein-
forcement, which is characteristic of the form of trial and error
known as discrimination learning, higher organisms are able to
emerge with one response successfully conditioned to one combina-
tion of stimuli and with a quite different response successfully
conditioned to another combination of stimuli containing many of
the components of the first, provided some of the elements are dif-
ferent. At bottom this discrimination is possible because the
afferent impulse s, which arises from the stimulus element Si when
occurring concurrently with the stimulus element Sg, is to some
extent different from s 3 , which arises from the same stimulus ele-
ment, S Jf when occurring concurrently with a different stimulus
element, S 3 . The physiological summation of the several compo-
nent reaction potentials characteristic of various stimulus patterns
which have many, and even most, of their stimulus elements in
common, accordingly may result in the evocation without con-
fusion of the distinctive reaction conditioned to each. Thus each
of the forty or so elementary speech sounds is a fairly distinctive
pattern made up of a “fundamental” physical vibration rate and
a particular combination of higher partials. Each of the thousands
of words of the better-developed languages consists of a temporally
patterned sequence of these elementary speech sounds, stops, and
so forth. In reading, each letter is a complex visual pattern, each
word is a complex pattern of these letter patterns, and each sentence
is a temporally patterned sequence of printed word patterns. In-
deed, it is impossible to think of a life situation which is not pat-
terned to a considerable extent. The limiting case of this kind of

PRINCIPLES OF BEHAVIOR

398

learning is that in which a stimulus compound is conditioned to
evoke a reaction while the several components when acting alone
are consistently extinguished.

A FORWARD GLANCE

The main concern of this work has been to isolate and present
the primary or basic principles or laws of behavior as they appear
in the current state of behavioral knowledge; at present there have
been isolated sixteen such principles. In so far as these principles
or postulates are sound and sufficient, it should be possible to
deduce from them an extensive logical hierarchy of secondary prin-
ciples which will exactly parallel all of the objectively observable
phenomena of the behavior of higher organisms; such a hierarchy
would constitute a systematic theory of all the social sciences. Con-
siderable progress has been made in this direction (/, 2, 3, 4, 5, 6,
7 > 8 > 9 > 10 > 11 > 12 y 13y H, 15, 16, 17, 18, 19, 20, 22, 23, 24, 25, 26,
27, 28, 29), though because of the limitations in available space
only a random sampling of some fifty or so secondary principles
(corollaries) is included in the present volume; these are given

chiefly for purposes of illustrating the meaning of the primary prin-
ciples.

As the systematization of the behavior sciences proceeds, some
of the principles put forward above as primary will be found to
yield false deductions and will therefore be abandoned; some will
be discarded as primary principles because found derivable from
other primary principles and consequently will be placed in the
group of secondary principles; others will be found partially defec-
tive and will require modification; finally, entirely new postulates
will need to be added. The primary principles presented in the
preceding pages have been formulated with the certainty of these
future developments fully in mind. A sharp and definite formula-
tion as, in many cases, been given principles despite admitted
doubt as to their precise validity. It is believed that a clear formu-
ation, even if later found incorrect, will ultimately lead more
quickly and easily to a correct formulation than will a pussyfoot-
ing statement which might be more difficult to convict of falsity.
The primary task of a science is the early and economical discovery
of its basic laws. In the view of the scientifically sophisticated,
to make an incorrect guess whose error is easily detected should be
no disgrace; scientific discovery is in part a trial-and-error process.

399

GENERAL SUMMARY AND CONCLUSIONS

and such a process cannot occur without erroneous as well as suc-
cessful trials. On the other hand, to employ a methodology by
which it is impossible readily to detect a mistake once made, or
deliberately to hide a possible mistake behind weasel words, philo-
sophical fog, and anthropomorphic prejudice, slows the trial-and-
error process, and so retards scientific progress.

It is to be hoped that as the years go by, systematic treatises
on the different aspects of the behavior sciences will appear. One
of the first of these would naturally present a general theory of
individual behavior; another, a general theory of social behavior.
In the elaboration of various subdivisions and combinations of
these volumes there would develop a systematic series of theoretical
works dealing with different specialized aspects of mammalian be-
havior, particularly the behavior of human organisms. Such a
development would include volumes devoted to the theory of skills
and their acquisition; of communicational symbolism or language
(semantics) ; of the use of symbolism in individual problem solu-
tion involving thought and reasoning; of social or ritualistic sym-
bolism; of economic values and valuation; of moral values and
valuation; of aesthetic values and valuation; of familial behavior;
of individual adaptive efficiency (intelligence) ; of the formal edu-
cative processes; of psychogenic disorders; of social control and
delinquency; of character- and personality; of culture and accul-
turation; of magic and religious practices; of custom, law, and
jurisprudence; of politics and government; and of many other
specialized behavior fields.

As a culmination of the whole* there would finally appear a
work consisting chiefly of mathematics and mathematical logic.
This would set out with a list of undefined terms or signs whose
referents are publicly available to the observation of all normal
persons; such terms, because they can be directly conditioned to
the referents by differential reinforcement, should have a minimum
of ambiguity. From these undefined notions would be synthesized
by the incomparable technique of symbolic logic all the critical
concepts required by the system, for correct primary concepts are
just as important for valid systematization in science as are correct
primary principles; this should yield a complete set of wholly un-
ambiguous terms. From these terms or signs would be formulated
precise mathematical statements of the several postulates or pri-
mary molar principles which survive the intervening winnowing
process, together with such other principles as it may be found

PRINCIPLES OF BEHAVIOR

400

necessary to introduce; from these, by means of rigorous mathe-
matical processes, would be derived theorems paralleling all the
empirical ramifications of the so-called social sciences. Also there
should be derivable large numbers of theorems concerning the out-
come of situations never yet investigated; this latter group would
make possible practical behavior applications and social inventions.

If one may judge by the history of the older sciences, it will
be a long time before the “social” sciences attain a status closely
approximating that contemplated here. Nevertheless there is rea-
son to hope that the next hundred years will see an unprecedented
development in this field. One reason for optimism in this respect
lies in the increasing tendency, at least among Americans, to regard
the “social” or behavioral sciences as genuine natural sciences
rather than as Geistesvrissenschaft. Closely allied to this tendency
is the growing practice of excluding theological, folk, and anthropo-
morphic considerations from the list of the presumptive primary
behavioral explanatory factors. Wholly congruent with these ten-
dencies is the expanding recognition of the desirability in the
behavior sciences of explicit and exact systematic formulation,
with empirical verification at every possible point. If these three
tendencies continue to increase, as seems likely, there is good reason
to hope that the behavioral sciences will presently display a devel-
opment comparable to that manifested by the physical sciences in
the age of Copernicus, Kepler, Galileo, and Newton.

But we should not deceive ourselves. The task of systemati-
cally developing the behavior sciences will be both arduous and
exacting, and many radical changes must occur. Behavior scien-
tists must not only learn to read mathematics unde- standingly —
they must learn to think in terms of equations and the higher
mathematics. The so-called social sciences will no longer be a
division of belles lettres ; anthropomorphic intuition and a brilliant
6tyle, desirable as they are, will no longer suffice as in the days
of William James and James Horton Cooley. Progress in this
new era will consist in the laborious writing, one by one, of hun-
dreds of equations; in the experimental determination, one by one,
of hundreds of the empirical constants contained in the equations;
in the devising of practically usable units in which to measure the
quantities expressed by the equations; in the objective definition
of hundreds of symbols appearing in the equations; in the rigorous
deduction, one by one, of thousands of theorems and corollaries
from the primary definitions and equations; in the meticulous per-

GENERAL SUMMARY AND CONCLUSIONS 4 01

fonnance of thousands of critical quantitative experiments and
field investigations designed with imagination, sagacity, and daring
to test simultaneously the validity of both the theorems and the
primary principles and concepts from which the former have been
derived; in the ruthless discard or revision of once promising pri-
mary principles or concepts which have failed wholly or in part
to meet the test- of empirical validation.

There will be encountered vituperative opposition from those
who cannot or will not think in terms of mathematics, from those
who prefer to have their scientific pictures artistically out of focus,
from those who are apprehensive of the ultimate exposure of cer-
tain personally cherished superstitions and magical practices, and
from those who are associated with institutions whose vested inter-
ests may be fancied as endangered.

This great task can be no more than begun by the present
generation of workers. Hope lies, as always, in the oncoming
youth, those now in training and those to be trained in the future.
Upon them rests the burden of the grinding and often thankless
labor involved, and to them must rightfully go the thrill of intel-
lectual adventure and the credit for scientific achievement. Per-
haps they will have the satisfaction of creating a new and better
world, one in which, among other things, there will be a really
effective and universal moral education. The present work is pri-
marily addressed to them.

REFERENCES

1. Ford, C. S. Culture and human behavior. Scientific Monthly, 1942, 65,

546-557.

2. Gibbon, E. J. A systematic application of the concepts of generalization

and differentiation to verbal learning. Psychol. Rev., 1940, 47, 196-229.

3. Guthrie, E. R. The. psychology of learning. New York: Harper and

4. Guthrie. E. R. The psychology of human conflict. New York: Harper

and Bros., 1938. ... , • a

5. Hill, C. J. Goal gradient, anticipation, and perseveration m compound

trial-and-error learning. J. Exper. Psychol., 1930, 26, 566-585.

6. Holt, E. B. Animal drive and the learning process. New York: Henry

Holt and Co., 1931. . . . . .

7. Hull, C. L. Simple trial-and-error learning: a study in psychological

theory. Psychol. Rev., 1930, 37, 241-256.

8. Hull, C. L. Knowledge and purpose as habit mechanisms. Psychol.

Rev., 1930, S7, 511-525.

9. Hull, C. L. Goal attraction and directing ideas conceived as habit

phenomena. Psychol. Rev., 1931, S8, 487-506

402 PRINCIPLES OF BEHAVIOR

10. Hull, C. L. The goal gradient hypothesis and maze learning. Psychol .

Rev., 1932, 39, 25-43.

11. Hull, C. L. The concept of vhe habit-family hierarchy and maze learn-

ing. Psychol. Rev., 1934, 1,1, 33-52; 134-152.

12. Hull, C. L. The mechanism of the assembly of behavior segments in

novel combinations suitable for problem solution. Psychol. Rev ^
1935, 42, 219-245.

13. Hull, C. L. Mind, mechanism, and adaptive behavior. Psychol. Rev.,

1937, 44 , 1-32. f

14. Hull, C. L. The goal-gradient hypothesis applied to some ‘field-force

problems in the behavior of young children. Psychol. Rev., 1938, 45,
271-299.

15. Hull, C. L. Simple trial-and-error learning — an empirical investigation.

J . Comp. Psychol., 1939, 27 , 233-258.

16. Hull, C. L. Psychology seminar memoranda, 1939-1940. Bound mimeo-

graphed material on file in the libraries of the University of Iowa,
Oberlin College, Yale University.

17. Hull, C. L. Conditioning: outline of a systematic theory of learning.

Chapt. II in The psychology of learning, 41st Yearbook Natl. Soc.
Study of Education, Part II. Bloomington, 111.: Public School Pub.
Co., 1942.

18. Hull, C. L., Hovland, C. I., Ross, R. T., Hall, M., Perkins, D. T., and

Fitch, F. B. M athematico-deductive theory of rote learning. New
Haven: Yale Univ. Press, 1940.

19. Hull, C. L., and Mowrer, O. H. Hull's psychological seminars, 1936-38,

Notices and abstracts of proceedings. Bound mimeographed material
on file in the libraries of the University of Chicago, University of North
Carolina, and Yale University.

20. Miller, N. E., and Dollard, J. Social learning and imitation. New

Haven: Yale Univ. Press, 1941.

21. Reiche, F. The quantum theory. New York: E. P. Dutton, 1930.

22. Rouse, R. O. The oscillation function in compound trial-and-error learn-

ing. J. Comp. Psychol., 1943, 35, 177-186.

23. Spence, K. W. The nature of discrimination learning in animals. Psychol.

Rev., 1936, 43, 427-449.

24. Spence, K. W. The differential response in animals to stimuli varying

within a single dimension. Psychol. Rev., 1937, 44, 430-444.

25. Spence, K. W. Analysis of formation of visual discrimination habits in

chimpanzee. J. Comp. Psychol., 1937, 23, 77-100.

26. Spence, K. W. Continuous versus non-continuous interpretations of dis-

crimination learning. Psychol. Rev., 1940, 47, 271-288.

27. Spence, K. W. Failure of transposition in size-discrimination of chim-

panzees. Amer. J. Psychol., 1941, 64, 223-229.

28. Spence, K. W. The basis of solution by chimpanzees of the intermediate

size problem. J. Exper. Psychol., 1942, 31, 257-271.

29. Whiting, J. W. M. Becoming a Kwoma ; Leaching and learning in a

New Guinea tribe . New Haven: Yale Univ. Press, 1941.

GLOSSARY OF SYMBOLS

Note: The literal signs are arranged primarily in the alphabetical order
of the major letter constituting the sign, and secondarily according to the
subscripts. The non-literal signs are grouped at the end of the list.

• _

A = amplitude, magnitude, or intensity of a reaction; A = h'sE R —

a — empirical exponential constant in the equation expressing the
generalized or effective habit strength as a function of sHr and
d, i.e.,

Ji K = s HhC-^.

o' = empirical constant in the equation, $i t R =

sE R b ‘
cn

B = empirical constant in the equation, I R = ^ ^

V = empirical constant in the equation, stn

tPt

C D

temporal coincidence of a receptor impulse (§) and the beginning
of a reaction impulse (r).

empirical constant in the equation, I R = ^ *

conditions which produce the drive ( D ), the objective conditions-
from which D may be calculated.

D — strength of dominant primary drive operative in the primary
motivation to action after the formation of the habit involved.

D* = strength of primary drive (D) operative during
a habit.

the formation of

D = the joint strength of all the non-dominant drives active at a given
moment.

100

D + D

D + Mo

d = the number of j.n.d.'s lying between the two stimulus aggregates
b and S .

d' = D - D.

e ■= a mathematical constant properly having a value of 2.7183 but
here frequently given the value of 10 because more convenient
in use where logarithms are involved.

403

PRINCIPLES OF BEHAVIOR

404

& Er = excitatory potential, potentiality of reaction evocation; i.e.

F — y ^ + ^

s&r — Si + s d h r x

En =

S^R

D + M d

— — •

effective reaction potential, i.e., sEr = sEr — /«•

= momentary effective reaction potential; S E R as modified by ^) B »

F — the constant factor of reduction of the unrealized potential habit
strength under given learning conditions. Thus if at each rein-
forcement the unrealized potentiality of habit strength is reduced
by 1/10, F has a value of .1.

F ' =* amount of force.

/ = an unstated quantitative functional relationship, e.g.,

A = .141 sH r 4- 3.1 may be written, A = /(sHr), i-e*> A is a
function oi, s H B .

y = empirical constant in the equation, n = c' S E R — /'.

G — a need reduction or a stimulus which has been closely associated
with a need reduction; primary reinforcement; also a primary
goal reaction.

g = fractional portion of a goal reaction which may be split off from
G and carried forward in a behavior sequence as a fractional ante-
dating goal reaction.

sHr = habit strength conceived as a rough or approximate- stimulus-
response relationship to t H r .

sHr = the habit strength ($#«) which results from N reinforcements.

,Hr = habit strength conceived as a precise dynamic relationship be-
tween afferent and efferent neural impulses.

AsHft = increment of habit strength resulting from a single reinforcement.

sHr = effective habit strength : gH R — s HRe~ ya .

ssHr = Si+ 3 ^ r> i-e., the result of the summation of the habit strengths

associated with two or more stimulus elements or
aggregates.

h ' — number of hours, as of food privation.

h ' - empirical constant in the equation, A = k’ s Er ~ *'•

1 r = amount of reaction inhibition.

I R =* total amount of inhibitory potential, i.e., Ir — Ir+ s?r»
s Ir — amount of conditioned inhibitory potential.

GLOSSARY OF SYMBOLS 4°5

*"7 fi = amount of reactive inhibition t units of time after a given sequence
of reaction evocations, i.e.,

r "In = In x 10-«"\

A I B = the increment of reactive inhibition generated at a single reaction
evocation.

x = the exponential constant in a learning situation where s Hr =
m _ i The quantitative value of i is given by the equation,

*' - r-S •

= empirical constant in the equation, A = h' s Er —

j = empirical constant in the equation expressing the maximum habit
strength as limited by the delay in reinforcement (0 :

m' =

j ’ = empirical constant in the equation expressing the stimulus gener-
alization of habit strength,

sH r = sH

im.d. = discrimination threshold— the distance on a generalization con-
tinuum between two stimuli which, at the limit of practice, can be
reacted to differentially on 75 per cent of the trials.

k = empirical constant in the equation expressing the maximum habit
strength as limited by the quality and quantity of the reinforcing
agent employed per reinforcement, i.e.,

AT = M( 1 -

L = distance fle n gth) of movement.

3 Lr = reaction threshold, the minimal amount of effective reaction
potential ( s Er ) (the effect of oscillation being at a minimum)
that will mediate reaction evocation.

log = logarithm.

M = the physiological maximum of habit strength attainable under
optimal conditions. This value is taken as 100 habs.

Af 7 = the maximum of habit strength with unlimited practice as limited
by the amount and quality of the reinforcing agent, i.e.,

M' = Mil - 10-*”).

M d = the physiological maximum of drive (100 mots).

Mt - the physiological maximum of reactive inhibition (100 pavs).

406

PRINCIPLES OF BEHAVIOR

m =

the physiological maximum of habit strength, attainable with un-
limited practice, as limited by the asynchronism of S and R in the
reinforcement situation, e.g.,

t lit*

m = me

m* =

or m = m e .

the maximum of habit strength, with unlimited practice, as limited
by the delay in reinforcement, i.e.,

m' = M'e-*.

N = the number of reinforcements,
n =

the number of unreinforced reactions required to produce experi-
mental extinction, i.e.,

(B - W)( S E R - s L r )

n =

0 =

—

a function of sO« such that O = $0* wheD sOr < sH_r» but

O = s Hr when <P R > S H R .

the oscillatory weakening potentiality associated with effective
reaction potential ( S E R ).

P = coefficient of observable patterning, i.e., P = Q — Q.

P' =
V =

Q =
Q -
O' =
Q' =

Q =
R =

R «

Ru =

theoretical index of patterning, i.e., P* = 100^1 —
probability of reaction evocation.

per cent of empirical reaction evocations by the positive or rein-
forced phase of a stimulus compound.

per cent of empirical reaction evocations by the negative or non-
reinforced phase of a stimulus compound.

theoretical effective reaction potential of the negative portion of
a stimulus patterning situation.

theoretical effective reaction potential of the positive portion of a
stimulus patterning situation.

empirical exponential constant in the equation of the dissipation
of /*, «/* = 1 R X

(1) reaction or response in general (muscular, glandular, electri-
cal) ;

(2) more specifically, the reaction which occurs as the result of pre-
vious conditioning.

a reaction which is in the process of being conditioned to a
stimulus.

unconditioned response such as the flow of saliva in a Pavlovian
conditioned reflex experiment.

GLOSSARY OF SYMBOLS

407

R e = response to a conditioned stimulus before conditioning begins.

<g) = a response which is either unobservable or exceedingly feeble and
difficult of observation.

5=1. stimulus energy in general, e.g., the energy of sound, light, or
heat waves, pressure, etc.

2. more specifically, stimulus energy which evokes a response on
the basis of a previously formed habit.

S A = stimulation arising from apparatus employed in an experiment.

S D = drive stimulus, i.e., stimulation arising from a condition of need
or disequilibrium.

S e = conditioned stimulus such as the buzzer in a Pavlovian condi-
tioned-reflex experiment.

S u = unconditioned stimulus such as the food of the Pavlovian condi-
tioned-reflex experiment.

5 = a stimulus when considered as in the process of being conditioned
to a reaction.

a = afferent neural impulse resulting from the action of a stimulus
energy on a receptor, as 5 — ► s.

$d = drive stimulus receptor discharge.

8 C = afferent impulse arising from the action of a conditioned stimulus
on a receptor.

9a — afferent impulse. arising from the action on a receptor of a stimulus
energy arising from an apparatus employed in a learning situation.

h = an afferent neural impulse when considered as in the process of
being conditioned to a reaction.

5 = an afferent neural impulse as modified by afferent neural inter-
action.

► r ) or A (5 >22) = increment to a receptor-effector connection.

T = the time at which an instantaneous event occurs.

T c = the time of the beginning of the R of a ; C , .

Tr = the time of the beginning of a reaction.

T‘ s = the time of the beginning of a stimulus which is in the process of
being conditioned.

To = the time at which a reinforcement occurs.

( = (1) time in the sense of duration;

(2) the duration of the delay in reinforcement, i.e., t = Tc — T a .

408

PRINCIPLES OF BEHAVIOR

t' = T r — Ts — -66, when S acts continuously and overlaps the begin-
ning of R, where T R and T$ are given in seconds.

•

= T r — T s — .44, when S and R are practically instantaneous,
where T R and T$ are given in seconds.

tf" = the duration in minutes following a sequence of unreinforced
evocations of R during which neither reinforced nor unreinforced
evocations of R have occurred.

sf-R = the latency of a reaction evocation, the time intervening between
the beginning of the stimulus and the beginning of the response.

sU r — unlearned or native receptor-effector reaction potential.

sU r = momentary unlearned reaction potential.

u = empirical exponential constant in the equation expressing the
maximum habit strength as limited by the time a stimulus (5)
has been continuously acting when R occurs, i.e.,

m = m'e-*’.

v = empirical exponential constant in the equation expressing the
maximum habit strength attainable with unlimited reinforcement
as limited by the degree of S — R asynchronism, i.e.,

m = m'e-*”,

W = the amount of work, i.e., W = F'L.

xd = the magnitude of a reinforcing agent employed as in the equation,

M' = Af (I - e~ ku >).

A = increment, e.g., A S H R .

c = used in ^mathematical logic and read as “is," e.g., xeS is read,
x is

2 — the sum of a series, as 2A S H R , which means the sum of the incre-
ments of a habit strength resulting from a series of reinforcements.

c s°r ~ tbe standard deviation of the oscillation of reaction potential.

4>(t) = the standard probability function.

fa = ) <t>(t)dt.

J OO

= a sign used in mathematical logic meaning “and."

ZD = a sign of implication used in mathematical logic and read, “if . . •»
then, * ; e.g., x3 y is read, “If x, then y, " i.e., x implies y.

GLOSSARY OF SYMBOLS

4°9

££ = a sign used in mathematical logic, e.g., Ox), and read, “There is
an x such that • • •

4* = physiological summation, e.g., + sHr 1 —

bHr 1 + sHr? —

sHri X sH

100

> = greater than, e.g., 5 > 4.

< = less than, e.g., 4 < 5.

— > «= a causal receptor-effector relationship inherited or at least in
functional condition at the outset of a learning situation.

> = a causal receptor-effector relationship which is acquired by the

organism.

= a causal relationship other than that of a receptor-effector connec-
tion.

INDEX OF NAMES

(Page numbers in bold-face type refer to the lists of references at the ends
of the chapters.)

Adrian, E. D., 41, 42, 49, 182
Allport, G., 101

Anderson, A. C., 148, 149, 151, 153,
155, 156, 163

Anrep, G. V., 49, 56, 83, 101, 123, 182,
203, 256, 263, 276, 303, 380
Arakelian, P., 235, 256, 276

Bass, M. J., 263, 276

Beach, F. A., 231, 237, 241, 256

Bechterev, V. M., 76, 83

Birge, J. S., 203

Blair, E. A., 309, 310, 321

Boeke, J., 51

Boring, E. G., 348

Bridgman, P. W., 30, 31

Brown, W., 321

Brunswik, E., 375, 380

Bugelski, R., 88, 91, 101, 106

Cajori, F., 15

Calvin, J. S., 290, 291, 296, 297, 302
Cannon, W. B., 68, 83
Cooley, J. H., 400
Copernicus, N., 400
Cowles, J. T., 89, 90, 91, 101
Crutchfield, R. S., 280, 294, 302

Darwin, C., 17, 31
Dashiell, J. F., 83
Day, A. S., 200, 223, 224
DeCamp, J. E., 152, 163
Dollard, J., 303, 402
Driesch, H., 23, 28, 31

Einstein, A., 20

Elliott, M. H., 231, 247, 248, 256
Ellson, D. G., 82, 235, 268, 269, 270,
271, 276, 285, 286, 302
Erlanger, J., 309, 310, 321
Euclid, 9
Eurman, 368

Fechner, G. T., 323
Feokritova, J. P., 173, 174
Finan, J. L., 81, 83
Finch, G., 231, 256

Fisher, R. A., 181

Fitch, F. B., IS, 31, 49, 123, 303, 321,
402

Fitts, P. M., 287, 302
Fletcher, F. M., 132, 134
Fogelsanger, H. M., 219, 225
Ford, C. S., 401
Freud, S., 101, 241, 252
Frolov, G. P., 84, 85, 86, 89, 90, 91
Fulton, J. F., 49, 51, 52, 54, 56, 348

Galileo, 11, 19, 20, 400
Gallagher, T. F., 256
Gantt, W. H., 124, 125, 126, 127, 128,
129, 134, 181, 182
Gengerelli, J. A., 294, 303
Gibson, E. J., 348, 401
Graham, C. H., 40, 41, 182
Grice, G. R., 152, 156, 163, 293, 294,
303

Grindley, O. C., 91, 92, 94, 101, 125,
126, 127, 129, 134, 142
Guthrie, E. R., 25, 31, 172, 182, 191,
203, 401

Haas, E. L., 136, 137, 140, 141, 164
Hall, M., 15, 31, 49, 123, 303, 321, 402
Hamilton, E. L., 136, 137, 138, 140,
141, 163

Hays, R., 304, 305, 306
Heathers, G. L., 235, 256
Heron, W. T., 237, 257
Hilgard, E. R., 83, 203, 290, 303, 331,
332 348

Hill, C. J., 306, 321, 324, 345, 348, 401
Hipparchus, 7
Holt, E. B., 401

Hovland, C. I., 15, 31, 49, 103, 112,
113, 123, 181, 184, 185, 186, 200, 201,
203, 236, 256, 260, 261, 263, 264, 265,
266, 275, 276, 289, 291, 292, 293, 303,
321, 339, 347, 348, 402
Hudgins, C. V., 291, 303
Hull, B. I., 107, 108
Hull, C. L„ 15, 31, 49, 83, 101, 123,
163, 182, 203, 224, 233, 234, 250, 251,
256, 263, 276, 303, 321, 380, 401, 402

411

INDEX OF NAMES

412

Humphrey, G., 372, 374, 380
Humphreys, L. G., 203, 337, 348

•

James, W., 400

James, W. T., 299, 303

Jenkins, T. N., 257

Jenkins, W. O., 162, 163, 302, 303

Jones, H. M., 278, 279, 280, 300, 303

Kaplon, M. D., 127, 128, 134
Kappauf, W. E., 165, 166, 167, 168,
169, 171, 182, 207
Kelley, T. L., 163
Kepler, J., 400
Koffka, K., 31, 380
Kohler, W., 48, 49, 219, 225, 380

Lashley, K. S., 189, 203, 218, 225
Leeper, R., 234, 235, 250, 251, 256
Lorente de N6, R., 42 *

Lumsdaine, A. A., 192, 193

McCurdy, H. G., 257
Marquis, D. G., 83, 203, 290, 303, 331,
332, 348

Miles, W. R., 143, 163, 237, 256
Miller, N. E., 143, 163, 237, 256, 278,
281, 289, 297, 298, 303, 402
Moore, C. R., 231, 256
Mowrer, O. H., 101, 278, 279, 280, 281,
289, 297, 298, 300, 303, 402
Murchison, C., 49
Murphy, E., 83
Murphy, G., 83
Murphy, W., 83
Murray, H. A., 315, 321

Newton, I., 6, 7. 8, 11, 15, 19, 20, 400
Nikiforovsky, M. P., 236

Pavlov, I. P., 20, 42, 47, 48, 49, 50, 56,
58, 75, 76, 78, 79, 80, 81, 83, 84, 87,
92, 93, 94, 97, 99, 101, 110, 123, 165,
169, 171, 173, 182, 184, 203, 204, 208,
211, 217, 220, 221, 222, 225, 232, 236,
249, 256, 259, 260, 261, 263, 264, 268,
269, 270, 272, 273, 276, 280, 282, 283,
284, 288, 290, 296, 303, 350, 351, 352,
353, 354, 368, 376, 380
Perin, C. T., 67, 101, 106, 123, 139,
140, 142, 143, 158, 161, 163, 181, 182,

227, 228, 229, 231, 232, 236, 242, 244,
247, 254, 257, 325, 326, 347, 348
Perkins, D. T., 15, 31, 49, 123, 222,
223, 303, 321, 402
Planck, M., 20
Price, D., 256

Razran, G. H. S., 376, 380
Reiche, F., 402
Richter, C. P., 62, 63, 64, 67
Rietz, H. L., 321
Robinson, E. S., 303
Rosenblueth, A., 42, 43, 49, 182
Ross, R. T., 15, 31, 49, 123, 303, 321,
402

Rouse, R. O., 175, 402
Russell, B., 15

Schlosberg, H., 165, 166, 167, 168, 169,
171, 182, 207
Shepard, J. F., 219, 225
Shipley, W. C., 192, 193, 203
Simley, O. A., 104, 105, 123, 336, 337,
346 348

Skinner, B. F., 82, 87, 88, 89, 92, 101,
106, 139, 227, 230, 231, 235, 236, 237,
241, 247, 257, 279, 287, 302, 304
Spearman, C., 320, 321
Spence, K. W., 101, 267, 276, 402
Spinoza, B., 8, 15
Stirling, A. H., 15
Stone, C. P., 231, 257
Switzer, S. A., 182, 236, 257, 291, 303

Thomson, G. H., 321
Thorndike, E. L., 71, 78, 80, 81, 83,
135, 136, 137, 159, 160, 162, 163, 164,
191, 203, 302, 303, 306, 321
Thurstone, L. L., 334, 348
Tolman, E. C., 31
Tsai, L. S., 294, 303

Wada, T., 61, 62, 67
Wang, G. H., 63, 67
Ward, L. B., 296, 303
Warden, C. J., 136, 137, 140, 141, 164,
241, 257

Warner, L. H., 176, 182, 257
Washburn, M. F., 135, 137, 159, 160,
164

Waters, R. H., 294, 303

Watson, J. B., 31, 136, 137, 140, 164

Weber, E. H., 154, 155, 156, 323

INDEX OF NAMES

4*3

Weiss, P., 45, 49, 310, 321
Wendt, G. R., 83
Wertheimer, M., 379
Wheeler, R. H., 294, 303
White, W. H., 15
Whitehead, A. N., 15
Whiting, J. W. M-, 402
Williams, S. B., 106, 121, 123, 181, 227,
257, 338, 347, 348

Wolfe, J. B., 91, 101, 127, 128, 134,
137, 138, 140, 141, 142, 143, 158, 160,
161, 164

Wolfle, H. M, 170, 171, 172, 173, 182

Woodbury, C. B., 47, 49, 304, 305, 306,
351, 352, 353, 355, 356, 368, 369, 370,
375, 380

Woodrow, H., Ill, 123

Yarbrough, J. U., 176, 182
Yoshioka, J. G., 152, 153, 155, 160, 162,
164

Young, P. T., 67, 257
Youtz, R. E. P., 268, 276

Zavadsky, I. V., 272, 273
Zener, K. E., 208, 231, 257

Title

Author,

Accession No.

Call No.

Borrower’s

No.

Issue

Date

Issue

Date

INDEX OF SUBJECTS

Action: coordination of, 50 f.; deter-
mination of, 226 f.; habitual, 21; of
muscles, 50 f. ; problem of, 50 f.
Action potentials, 40 f.

Activity : of organism, 17 f.
Adaptation: afferent neural interac-
tion in, 385; contributions of evo-
lution, 385 f.; and learning, 386 f.;
perse verative stimulus trace in, 385 ;
receptors and effectors in, 384 f.; in
systematic behavior theory, 66
Adaptive behavior: and afferent neu-
ral interaction, 385; organic basis
of, 18 f.; of organisms, 384 f.; and
perseverative stimulus trace, 385
Afferent generalization continuum,
188 f.

Afferent neural impulse, 41 f.

Afferent neural interaction : and adap-
tive activity, 385; and behavioral
variability, 321; and configuration
psychologies, 48 f.; corollaries, 217,
219 f.; and patterning, 356 f., 359;
principle of, 216

After-discharge: of heart, 42; of optic
nerve, 40

All-or-none law : of muscle fiber.

51 f.

Amount of reinforcement hypothesis,

128; implications of, 129 f.
Amplitude of reaction, 339 f.
Antecedent conditions: example of, 4,

22, 26, 57, 102
Antedating goal reaction, 74
Anterior stimulus a synchronism gra-

dient, 172

Anticipatory goal reaction, 74
Acquired habit strength: of compo-
nents, 206 f.

Associative inhibition, 341
Asynchronism : gradients, 171 fj, 180;
point of optimal stimulus, 171, 180;
stimulus-response, 167 f.

Backward conditioning, 171 f.; Wolfle’s
experiment, 170 f.

Behavior: adaptive, 18 f., 25; future
science of, 398 f. ; innate, 57 f. \ mal-
adaptive, 25; molar, 25; motivated.

60 f.; objective theory of, 16 f.; or-
ganic basis of, 18 f.; purposive,
25 f.; sequence, 96

Behavior theory : antecedent condi-
tions, 382 f . ; consequent conditions,
382 f.; future of, 398 f.; objective
vs. subjective approach, 25 f.; sym-
bolic constructs in, 382 f. '

Behavioral oscillation, 304 f. (see also
Oscillation); asynchronism of, 308;
and concept of oscillatory force,
313 f.; effect of behavioral sciences,
316, 393; of effective habit strength,
318 f.; and effective reaction poten-
tial, 313 f., 393 f.; experimental

demonstration of, 304 f. ; Major
Corollary III, 319; mathematico-
deductive theory of rote learning,
320; and momentary effective re-
action potential, 313 f.; and normal
law of chance, 306, 310 f., 316 f.;
Postulate X, 319; and probability
of reaction evocation, 308 f.; “reac-
tion-evocation” paradox, 314 f. ; and
reaction threshold, 325; in simple
conditioning, 307 ; Spearman’s group
factor theory, 320 ; and spontaneous
neural discharge, 310; spontaneous
neural oscillation, 309 f.; submolar
causes of, 309 f.; as symbolic con-
struct, 393 ; Thorndike’s experiment,
306 f.; Weiss’s study, 310

Cathode ray oscillograph, 205
Compound conditioned stimuli: dis-
tribution of habit strength, 206 f.;
functional dynamics of, 204 f.; stim-
ulus complexity, 204 f.

Compound stimulus: aggregates,
209 f.; conditioned, 204 f.; corollar-
ies, 214 f., 217, 219 f.; and law of
primary reinforcement, 206; in
stimulus generalization, 190 f.
Concepts: of afferent generalization
continuum, 188 f.; of effective habit
strength, 181 f.; of stimulus dimen-
sion, 188 f.

Conditioned inhibitory potential, 392
corollary, 283; derivation of, 283

41 *

INDEX OF SUBJECTS

416

and generalization, 283 ; . Pavlov’s
study, 282 f.; Postulate VIII, 300

Conditioned reaction, 74 f. ; backward,
171 f.; cyclic-phase, 174 f.; Feokri-
tova’s experiment, 173 f.; trace,
173 f.

Conditioned reflex, 74 f. ; Bechterev
type of, 75; compared with selec-
tive learning, 76 f . ; “first order,”
85; “higher order,” 85, 90, 93 f.; as
learning reinforcement, 76 f., 386;
Pavlovian, 78 f. ; primary reinforce-
ment in, 75 f . ; “second order,” 85,
90; typical experiment, 75

Conditioned reflex learning: and
amount of reinforcing agent, 124 f.;
Gantt’s experiment, 124 f.

Conditioned stimulus: compound,
204 f.; habit strength and duration
of, 165 f.

Configuration, 44, 48 f.

Consequent conditions: in behavior
theory, 382 f.

Cyclic-phase conditioned reaction,
165 f.

Delay of reinforcement: affecting
preference for act, 146 f.; Ander-
son’s study, 148 f.; corollaries, 147 f.,
151 f., 154, 157; early attacks on
problem, 136 f.; and habit strength,
135 f.; Hamilton’s study, 136 f.;
origin of problem, 135 f.; Perm’s
experiment, 139 f., 161 f.; rate of
discrimination and, 152 f.; reconcili-
ation of experimental paradoxes,
140 f.; relation to Weber’s Law,
154 f.; Warden and Haas’ study,
140; Watson’s study, 136; Wolfe’s
experiment, 138, 160 f.; Yoshioka’s
study, 153 f.

Determination of action: Elliott’s
study, 231 f.; Penn’s study, 227 f.;
role of drive in, 226 f.; role of habit
strength in, 226 f. ; Stone’s experi-
ment, 231 ; Williams’ study, 227 f.

Differential reactions: corollary of,
251; Hull’s investigation, 233 f.; to
identical environmental situations,
233 f . ; Leeper’s study, 234 f.

Differential reinforcement: and nega-
tive patterning, 363 f.; and positive
patterning, 360 f.

Disinhibition: corollaries, 288 f., 293;
and curve of extinction, 291 f , ; of
extinction effects, 272 f.; Hovland’s
study, 291 f.; inhibition and, 392;
phenomenon of, 287 f.; simultane-
ous, 273; Switzer’s study, 291; Za-
vadsky’s study, 272 f.

Doctrine: of “emergentism,” 26 f.

Drives: aspect of motivation, 131; as
intervening variables, 57 f., 66 f.;
and need, 57 f.; Postulate VIII,
300; primary, 59 f.

Drugs: influence on experimental ex-
tinction, 236 f.; Miller and Miles
investigation, 237 ; Skinner and
Heron’s study, 237 ; Switzer’s study,
236

Effective habit strength: concept of,
187 f.; Major Corollary II, 253; os-
cillation of, 308 f., and probability
of reaction evocation, 308 f.; and
reaction-evocation power, 210 f.

Effective reaction potential : ampli-
tude of reaction and, 339 f. ; concept
of, 281 f.; corollaries, 284 f., 289 f.,
297; as function of reaction latency,
336 f.; incompatible reaction poten-
tials, 341; and inhibition, 277 f.;
Postulates XII, XIV, and XVI,
344; and reaction evocation, 326 f.;
and reaction potential, 226 f., 390 (.;
related to reaction-evocation prob-
ability, 326 f.; resistance to experi-
mental extinction, 337 f.

Effector: and adaptation, 384 f.; dis-
charge, 72

“Emergentism”: meaning of, 26 f.

Empirical reaction threshold: in con-
ditioning, 324 f.; Hill’s study, 324 f.;
individual demonstrations of, 324 f.

Entelechy, 258; Driesch’s, 23, 28

Environment: external, 16; inani-
mate, 16; internal, 16; organismic,
16

Equations: 1, 119; 2, 119; 3, 120; 4,
120; 5, 6, 7, 8, 9, 10, 121; 11, 134;

12, 134; 13, 160; 14, 162; 15, 163;

16, 178; 17, 18, 19, 20, 179; 21, 22,

23, 180 ; 24, 25, 26, 27, 28, 181; 29,

199; 30, 200; 31, 201 f.; 32, 223 ; 33,
34, 254 ; 35, 255 ; 36, 37, 38, 39, 40,
41, 42, 300; 43, 301; 44, 319; 45,

4*7

INDEX OF SUBJECTS

320 ; 46, 344 ; 47, 48, 49, 50, 51, 345;
62, 53, 355

Excitation gradient: interacting with
extinction gradient, 265 f.

Experimental extinction : corollaries,
287, 293; as corrective mechanism,
262; disinhibition and, 272 f., 291 f.;
Ellson’s study, 270 f., 275 f.; exam-
ples of, 259 f.; as function of effec-
tive reaction potential, 337 f., 347 ;
as function of unreinforced reac-
tions, 260 f.; habit strength and
resistance to, 106 f. ; Hovland’s
studies, 260 f., 263 f., 268 f., 270 f.,
275; influence of drugs, 236 f.; and
inhibition, 391 f.; interaction of gra-
dients, 265 f . ; motivational status,
391; Pavlov’s studies, 259 f., 263,
270 ; perseverational effects of,
236 f . ; Postulate XIV, 344 ; reaction
generalization of, 267 f.; and Remi-
niscence, 391 f.; as secondary phe-
nomenon of reactive inhibition,
277 f., 391 f.; spontaneous recovery
in, 269 f., 391 f.; stimulus generali-
zation of, 262 f.; and unadaptive
habits, 258 f.; Williams’ study,
106 f. ; Zavadsky’s study, 272 f.
~F.y tinp.tion, 88 (see also Experimental
extinction) ; experimental, 106 f.,
236 f., 258 f.; generalized, 101; of
secondary reinforcement, 90 f . ; 100 f.

“First order” conditioned reflex, 85
Fractional component: of goal reac-
tion, 100

Frequency: of impulses, 41 f.
Functional autonomy : as self-rein-
forcement, 101

Functional dynamics: of compound
conditioned stimuli, 204 f.

Galvanic skin reaction, 103 f.; Hov-
land’s study, 260 f.

Generalization : and behavioral varia-
bility, 321 ; of extinction effects,
262 f.; and positive patterning,
366 f.; response, 183; response in-
tensity, 316; stimulus, 183 f., 216 f.,
389 f. ; stimulus- response, 183
Goal, 25, 95; attainment, 26; gradi-
ent, 96, 100, 145; as reinforcing
state of affairs, 95 f.

Goal gradient hypothesis, 145; equa-
tion 14, 162; formulation of, 142 f.;
and gradient of reinforcement, 142;
and habit increment, 387 f.

Gradient: anterior stimulus asynchro-
nism, 172, 180; of excitation, 265 f.;
of extinction, 263 f.; of generaliza-
tion, 187 ; goal, 96, 100, 142 f 145,
160, 162; of habit strength, 173;
posterior - stimulus asynchronism,
171, 180; of reinforcement, 94, 143 f.,
159 f., 162, 388

Growth constant, 114, 119, 127, 129

Growth function: negative, 145; posi-
tive, 104, 332, 335

Hab: computation of, 119 f., 129; de-
fined, 114

Habit, 21, 74, 109; defined, 102; for-
mation, 102, 204 f.; increment,

387 f . ; in learning, 387 ; and rein-
forcement, 387; strength, 108 f.;
summation, 212 f.; unadaptive, 258 f.

Habit formation: amount of rein-
forcing" agent and curve of, 127 f.;
complexity of stimulus, 204 f.;
working hypothesis of, 128

Habit increment per reinforcement:
conditions influencing, 387 f.; goal
gradient, 388; gradient of reinforce-
ment, 388; primary reinforcing
agent, 388; secondary reinforce-
ment, 388; stimulus-response asyn-
chronism, 388

Habit strength : acquired, 206 f. ; con-
cept of, 108 f.; conclusion, 129; de-
lay and reinforcement and, 135 f.;
distribution of, 206 f.; duration of
conditioned stimulus and, 165 f.;
equations, 222 f.; as function of
amount and nature of reinforcing
agent, 124 f.; as function of num-
ber of reinforcements, 102 f., 112 f.,
387; how to compute, 119 f.; how
to compute increment of, 120; Kap-
pauf and Schlosberg’s study, 165 f.;
loading, 207 f . ; per cent correct re-
action evocation, 107 f.; Postulate
TV, 178; and primary motivation,
390; qualification as quantitative
scientific construct, 122; in quanti-
tative derivation of reaction poten-
tial, 242 f.; and reaction latency,
104 f.; and reaction magnitude.

INDEX OF SUBJECTS

418

103 f.; and reaction potential corol-
lary, 247 ; and resistance to experi-
mental extinction, 106 f.; and stim-
ulus drive corollaries, 248; stimulus
generalization, 389; stimulus -re-
sponse asynchronism, 165 f.; stimu-
lus trace, 169 f.; symbolic represen-
tation of, 111 f . ; theoretical curve
of growth, 114 f.; Wolfle’s experi-
ment, 170 f.

Habit summation: corollaries, 214 f.;
difficulty of applying equations,
223 f.; principle of, 212 f.

“Higher order” conditioned reaction,
85, 90; possibility of, 93 f.

Humphrey’s arpeggio paradox; confu-
sion about, 374; resolved, 372 f.

Hypothesis: Kauppauf - Schlosberg,
168; Mowrer-Miller, 278; of neural
interaction, 42 f.; of primary moti-
vation, 226 f.; of stimulus trace, 42

Incentive : and amount of reinforcing
agent, 131 f.; aspect of motivation,
131 ; concept of, 131 ; Fletcher’s ex-
periment on, 132; as secondary mo-
tivation, 226

Inhibition: associative, 341; condi-
tioned, 392; corollaries, 282 f. ; and
disinhibition, 392; and effective re-
action potential, 277 f. ; and experi-
mental extinction, 391 f.; motiva-
tional status, 391 f.; Postulates VIII
and IX, 300; reaction potential,
392 f.; reactive, 327. 391 f.; of rein-
forcement, 289 f.; reminiscence and,
391 f.; and spontaneous recovery,
391 f.

Inhibitory potential (see also Inhibi-
tion) : conditioning of, 281 f.; cor-
ollaries, 282 f., 288; and effective
reaction potential, 281 f.; as logical
construct, 278; Mowrer and Jones*
study, 279 f.; Pavlov’s study, 282 f.;
Postulates VIII and IX, 300; prin-
ciples related to, 277 f. ; problems
of conditioning of, 301 f.; quantita-
tive concept of, 277 f.; reiatiu— ;hip
of “work” to, 279 f.; its stimulus
generalization, 281 f.; unit of, 280 f.

Innate behavior: characteristics of,
57 f . ; and tendencies toward varia-
tion, 58 f.

Insight, 25

Intensity: of reaction, 339 f.

Intents, 25

Interaction: environmental - organ-
ismic, 16 f., 25 f.

Intervening variables! role of, 21 f.,
23, 30, 57 f.

jm.d., 190

Kinaesthesis, 35

Latency: of optic nerve fiber, 40; of
reaction, 104 f., 336 f.

Law: all-or-none, 51 f.; of “effect,”
78, 135 f.; of gravitation, 4 f., 11; of
“least action,” 294; of “less work,”
293 f.; of “minimal effort,” 294; of
primary reinforcement, 71, 206; of
probability, 160, 306, 310 f., 316 f.,
328; of “recency,” 135; of rein-
forcement, 98, 206, 258; of spon-
taneous recovery, 284; Weber’s,
154 f.

Learning: adaptation in, 386 f.; basic
curve of, 116; conditioned reflex a
special case of, 76 f., 386; discrimi-
nation, 265 f. ; equations of, 389;
equations fitted to curves of, 120 f.;
example of, 70‘f. ; general nature of,
68 f.; and habit, 387; and latency
of reaction, 110; and magnitude of
reaction, 110; primitive trial and
error, 386; and reinforcement,
386 f.; selective, 70 f., 76 f.; theo-
retical curve of, 117
Limen, 324

Logical constructs, 21 f., 23, 113; of
inhibitory potential, 276 f.; use of,

111

Magnitude: Hovland’s study, 103 f.;
of reaction, 103 f.

Major corollaries: I, 199; II, 253; III,
319; IV, 378; V, 379
Maladaptive behavior, 25
Mechanism: defined, 384
Molar analysis, 112
Molar behavior: contrasted with mo-
lecular behavior, 20 f. ; explanation
of, 17; task of, 19
Molecular behavior, 20 f.

Momentary effective reaction poten-
tial, 313 f.; Postulate XI, 344; Pos-
tulate XIII, 344; Postulate XV,

INDEX OF SUBJECTS

344; Postulate XVI, 344; mathe-
matical statement of postulates,
344 f.

Monotonic habit-reaction relation-
ship: corollary, 215; principle of,
212 f.

Mote: unit of strength of primary
drive, 238

Motivated activity, 60 f.; aspects of,
131; Richter’s studies on hunger
and sex drive, 62 f.; thirst, 60 f.;
Wada’s hunger study, 61 f.; Wang’s
sex study, 63

Motivation: and incentives, 226; pri-
mary, 226 f . ; secondary, 226
Motor end-plate, 51
Mowrer-Miller hypothesis: statement
of, 278; submolar principle from,
278 f.

Muscles* action of, 50 f.; all-or-none
law, 51 f.; unlearned coordination
of, 53 f.

Need: and drives, 57 (.; modal reac-
tions to, 59 f.; or organism, 17 f.;
reduction, 226

Negative growth function: goal gra-
dient, 145; and habit strength, 145
Negative patterning: Corollary IV,
365 ; derived, 363 f . ; by differential
reinforcement, 363 f. ; of simultane-
ous stimulus compounds, 363 f.
Neural interaction (see also Afferent
neural interaction) : and configura-
tion psychologies, 48 f . ; hypothesis
of, 42 f.; Pavlov’s statement of,
47 f.; Rosenblueth’s study, 42 f.
Neurological approach, 19 f.

Normal law of probability: and be-
havioral oscillation, 306, 310 f.; in
behavioral sciences, 316 f. ; and dis-
tribution of oscillatory force, 314;
equation of, 320; graphic represen-
tation, 312; and magnitude of mus-
cle contraction, 310; % the ogive, 313;
Postulate XII, 344

Objectivism: in behavior theory, 30;
vs. teleology, 24 f.

Operational definition: Bridgman, 30
Organisms: adaptive activity of,

384 f.; survival of, 32 f.; as self-
maintaining mechanisms, 384 f.

419

Oscillation, 45; corollary, 289; spon-
taneous, 149 f.

Pattern, 44

Patterning (see also Negative pat-
terning, Positive patterning, Si-
multaneous stimulus patterning.
Spontaneous stimulus patterning.
Stimulus patterns, and Temporal
stimulus patterning) : corollaries of,
358 f., 363, 365, 368, 371 f.; empirical
patterning coefficient, 355; equa-
tions of patterning coefficients, 355,
379; functional dynamics of, 374 f.,
396 f.; and Gestalt psychology,
379 f.; Humphrey’s arpeggio para-
dox resolved, 372 f.; Major Corol-
lary IV, 378; Major Corollary V,
379; negative, 350, 353, 363 f., 366 f.;
Pavlov’s work, 350 f.; physiological
summation in, 397 ; positive, 350,
352, 360 f.; principles involved in,
354 f.; simultaneous stimulus, 350 f.;
spontaneous, 356 f.; temporal stim-
ulus, 350, 368 f.; theoretical pat-
terning coefficient, ,355; Woodbury's
study, 351 f.

Pav : unit of inhibitory potential,
280 f.

Perseveration: Lorente de No’s study,
42; nature of, 41 f.; or reverbera-
tion, 42

Perserverative stimulus trace : and
adaptation, 385; and adaptive be-
havior, 385

Perseverative trace, 100, 71 ; and tem-
poral stimulus patterning, 371

Physiological: maximum, 114; sum-
mation, 102, 395 {., 397

Positive growth function, 229 f., 332,
335; basic principle of, 114; and
theoretical curve in habit-strength
formation, 114 f.

Positive patterning: Corollary III,
363; Corollary V, 368; with de-
creased generalization, 366 f.; de-
rived, 360 f.; by differential rein-
forcement, 360 f.; of simultaneous
stimulus compounds, 360 f.

Posterior stimulus asynchronism gra-
dient, 171

Postulates: as primary principles, 2 f.,
26; I, 47; II, 47; III, 66; IV, 178;
V, 199; VI and VII, 253; VIII and

420

INDEX OF

IX, 300; X, 319; XI, XII, XIII,
XHV, XV, and XVI, 344

Primaiy motivation: concept of re-
action potential, 239 f.; concept of
strength of primary drive, 238 f.;
differential reactions to identical
situations, 233 f.; drive as a factor
in, 226 f.; effect of sex hormones,
237 f.; and extinction, 391; habit
strength as factor in, 226 f., 390;
influence of drugs, 236 f.; Major
Corollary II, 253 f.; mediation of
behavior, 390; Postulates VI and
VII, 253 f. ; and reaction potential,
226 f., 390; stimulus-intensity gen-
eralization applied to drive, 235 f.;
as symbolic construct, 390; twelve
corollaries of, 247 f.

Primary reinforcement, 68 f.; in con-
ditioning, 76; critical factor in, 81;
Finan’s study, 81 f.; law of, 71, 206 f
learning, a process of, 71 ; in stimu-
lus compound, 206; and Thorn-
dike’s “law of effect,” 80.

Primary stimulus generalization: cor-
ollary, 219; principle of, 216 f.;
Shepard and Fogelsanger’s study,
219

Primary stimulus-intensity generali-
zation: applied to drive stimulus,

235 f.

Principle: of afferent neural interac-
tion, 47 f., 216 f.; and axioms, 2; of
habit summation, 212 f.; molar, 20;
molecular, 20; of monotonic habit-
reaction relationships, 212 f.; as
postulates, 2; primary, 2 f., 25; of
primary stimulus-intensity generali-
zation, 216 f., 235 f.; secondary, 2 f.,
25; of stimulation, 39

Probability, 262

Quasi-patterning. See Spontaneous
patterning

Reaction: amplitude, 339 f., 344, 347;
differential, 233 f.; effect of drugs,

236 f. ; evocation, 107 f.; galvanic
skin, 103 f.; generalization, 267 f.;
intensity, 339 f.; latency, 104 f.,
336 f., 344, 346; magnitude, 103 f.;
potential, 226 f.; threshold, 322 f.,
394

SUBJECTS

Reaction evocation : Bertha Iutzi
Hull’s study on habit strength and
per cent of, 107 f.; and compound
stimulus aggregates, 209 f.; corollar-
ies, 328, 333 f.; derivation of, 328 f.;
effective habit strength and, 210 f.;
and effective reaction potential,
326 f.; learning curves of, 330 f.,
346; paradox of, 314 f.; probability
of, 308 f., 326 f.; and reaction po-
tential, 394 f.; and reaction thresh-
old, 326 f., 394 f.

Reaction-evocation probability: and
behavioral oscillation, 308 f.; deri-
vation of, 328 f.; relation to effec-
tive reaction potential, 326 (., 394 f.
Reaction generalization: El Ison’s
study, 268 f.; of extinction effects,
267 f.

Reaction potential : corollaries, 247 f.,
283 f., 287, 289, 296; definition of,
239 f.; effective, 281 f., 392 f.; and
forgetting, 395; incompatible, 341;
and inhibition, 392 f.; Major Cor-
ollary II, 253; Postulate VII, 253 f.;
Postulate VIII, 300; and primary
motivation, 226 f., 390; as primary
motivational concept, 238 Ui quan-
titative derivation of, 242 f.; and
reaction threshold, 394 f.; response
evocation, 394 f.; successive extinc-
tion of same, 286 f.; units of, 239
Reaction threshold: in conditioning,
324 f.; chronuxie determinations,
324; defined, 324, 394 f.; empirical,
324; Hill’s study, 324 f.; and num-
ber of reinforcements, 394 f.; oscil-
lation components of, 345 f.; and
oscillation function, 325 f.; postu-
lates of, 344; reaction potential,
394 f.; and response evocation,
322 f., 394 f. ; true or initial, 325
Reactive inhibition, 327, 391 f.; ex-
perimental extinction arising from,
277 f.; and Mowrer-Miller hypoth-
esis, 278

Receptor: and adaptation, 384 f.;
analysis of environmental energies,
40 f.; discharge, 40 f., 71 f., 206; spe-
cialized, 18; types of, 33 f.
Receptor-effector connections: acqui-
sition of new, 68 f., 84 f.; in condi-
tioning, 76; innate, 69 f.; new, 73 f.;
strengthening of innate, 68 f., 72

421

INDEX OF SUBJECTS

Receptor-effector convergence, 184,
192

Reflex: chains, 55 f.; heterogenenous,
268; homogeneous, 268; stepping,
54 f.

Reification, 28

Reinforcement : corollaries, 291 , 296 f . ;
differential, 360 f., 363 f.; distrib-
uted, 295 f.; gradient of, 94; habit
formation and delay of, 135 f.;
and habit increment, 387 f.; habit
strength and number of, 112f., 387;
inhibition of, 260 f., 289 f.; law of,
98; and learning, 386 f.; primary,
68 f.; secondary, 84 f., 387

Reinforcing agent: Cowles’ study,
89 f.; Grindley’s experiment, 91 f.;
habit strength as a function of na-
ture and amount of, 124 f.; food
reward as, 98 f.; possible identity
of processes, 99 f.; primary, 89, 388;
problem of incentive and amount
of, 131 f., 134; secondary, 85, 86,
89 f., 99

Reinforcing factor: onset or termina-
tion of need-receptor impulse, 82 f.

Reminiscence: Calvin’s study, 296;
corollaries, 296; and distributed
reinforcements, 295 f.; and extinc-
tion, 391 f.; inhibition and, 391 f.;
phenomenon of, 295 f.

Reverberation, 42 ; of neural im-
pulses, 41 f.; and perseveration,
41 f.

Reward: as incentive, 131

Robot: use of, 27

Science: empirical aspects, If.; theo-
retical aspects, 1 f.

Scientific theory: anthropomorphism
in, 382; aspects of science, 1 f., 381 ;
constituting a logical hierarchy, 5 f.,
381 f.; deductive nature of, 2f.;
definition of, 2f.; differing from ar-
gumentation, 7f.; the nature of,
1 f., 381 f.; Newton’s system, 7; and
probability, 10, 12; and sampling,
10, 12; substantiation of postulates,
12 {.; theoretical and empirical con-
tributions, 9f.; “truth” status, 13 f.

“Second order” conditioned reaction,
85, 90

Secondary reinforcement: Bugelski’s
study, 88; differential causal effi-

cacy of, 89 f.; existence of, 84 f.;
extinction of, 90 f.; Frolov’s study,
84 f.; and habit increment, 387 f.;
problems concerning, 86; reactions
subject to, 87 f.; role in compound
selective learning, 95 f., 387; Skin-
ner’s study, 87 f.

Secondary stimulus generalization,
191 f.

Selective learning, 70 f., 77; amount
of reinforcing agent affecting rate
of, 125 f.; compared with condi-
tioned-reflex learning, 76 f.; Grind-
ley’s experiment, 125 f.; Wolfe and
Kaplon’s study, 127

Sensitization, 211

Sex hormones: Beach’s survey, 237 f.;
effect on motivation,' 231, 237 f.;
Stone’s study, 231

Sham feeding, 99

Simple discrimination learning: re-
sulting from interaction of gradi-
ents, 265 f.

Simultaneous stimulus patterning
(.see also Patterning) : experimental
examples of, 350 f.; Major Corol-
lary IV, 378; Pavlov’s work, 350 f. ;
spontaneous, 356 f.; Woodbury’s
study, 351 f.

Specialized receptors, 18

Spontaneous discharge, 59

Spontaneous emission: of neural im-
pulses, 44 f., 310; Weiss’s investiga-
tion, 45, 310

Spontaneous oscillation: Blair and
Erlanger’s investigation, 309 f.; of
neural conductors, 309 f.; as varia-
bility in habit strength, 149 f.

Spontaneous recovery: corollaries,
284 f., 287 ; Ellson’s study, 270 f.,
275 f.; of extinction effects, 269 f.,
391 f.; incompleteness of, 284 f.;
and inhibition, 391 f.; Pavlov’s
study, 270

Spontaneous stimulus patterning:
Corollary I, 358 f.; Corollary II,
359; derived, 356 f.; with increased
neural interaction, 359

Stimulation: principle of, 39

Stimulus: aggregate, 207, 209 f., 349 f.;
analysis of environmental energies,
40 f. ; complexity in typical condi-
tioning, 204 f.; compound, 190 f.,
395 f. ( see also Patterning); dimen-

4 22

INDEX OF SUBJECTS

sion, 188 f.; element, 207, 349 f.;
generalization, 183 f., 216 f., 219;
nature of, 32 f.; trace, 42
Stimulus compounds ( see also Pat-
terning): dynamics of, 395 f.; physi-
ological summation in, 395 f.
Stimulus dimension: concept of
188 f.; Postulate V, 199
Stimulus-evocation paradox, 194 f.;
its resolution, 196

Stimulus generalization: of condi-
tioned inhibition, 281 f.; corollaries,
283, 285 f.; dimensions of, 389 f.; of-
extinction effects, 262 f.; and habit
strength, 389; Hovland’s studies,
184 f., 186 f., 200 f.; incompleteness
of, 284 f.; Lumsdaine’s experiment,
192 f.; Major Corollary I, 199; by
means of identical stimulus com-
ponents, 190 f.; mediation of behav-
ior, 389; Postulate V, 199; primary
stimulus intensity, 186 f.; primary
stimulus quality, 184 f.; principle
of, 216 f.; secondary, 191 f; Ship-
ley’s experiment, 192; units of
measurement, 189 f.

Stimulus patterns (see also Pattern-
ing) : “calculus” of adaptive prob-
ability, 375; compound trial and
error, 375; functional dynamics of,
374 f.; serial trial and error, 376
Stimulus reception, 32 f.; of move-
ment, 35 f. ; of spatial relationships,
36 f.; of temporal relationships, 39
Stimulus-response asynchronism: an-
terior gradient, 172, 180; five cor-
ollaries of, 168 f.; and habit incre-
ment, 388; and habit strength,
165 f.; a neurological hypothesis of,
167 f.; posterior gradient, 171, 180
Stimulus trace, 42; perseverative, 385
Strength of drive stimulus: corol-
laries of, 247 f.; Heathers and
Arakelian’s study, 235; Major Cor-
ollary II, 253; physiological inter-
pretation, 240 f.; Postulate VI,
253 f.; as primary motivational con-
cept, 238 f.; in quantitative deter-
mination of reaction potential,
242 f.; stimulus-intensity generali-

zation applied to, 235 f.; unit of,
238

Strivings, 25
Subgoals, 90

Subjectivism, 27; in behavior theory,
30; prophylaxis against, 27 f.
Survival : of organism, 17, 32 f.
Symbolic constructs: behavioral os-
cillation, 393; in behavior theory,
282 f.; habit strength, 102 f.; pri-
mary motivation, 390; reaction
potential, 239 f.; strength of pri-
mary drive, 238 f.; summary of,
383

Symbolic representation: of habit
strength, 111 f.

Symbols, 21; in conditioned-reflex
literature, 75

Teleology: vs. objectivism, 24 f., 26
Temporal stimulus patterning (see
also Patterning) : corollaries, 371 f.;
Eurman’s study, 368; experimental
examples of, 368 f.; Major Corol-
lary V, 379; and perseverative
stimulus traces, 371 ; theoretical
analysis of, 370 f.; Woodbury’s in-
vestigation, 368 f.

Theorems: first and second order,
6f.; as secondary principles, 2
Theory : of behavior, 16 f.

Threshold, 323 f.

Trace: conditioned reaction, 173 f*
perseverative, 71, 100, 371
Trial and error: learning, 386 f.

Unadaptive habits: and experimen-
tal extinction, 258 f.; and generali-
zation in extinction of, 262 f.

Value, 25
Vitalism, 26

Wat: unit of reaction potential, 239
Work: corollary, 294 f.; Grice’s study,
293 f.; law of less work, 293 f.«
Mowrer and Jones’ study, 279 f •
Postulate VIII, 300; relationship to
inhibitory potential, 279 f.

flLLAHA IQBAL LIBRARY

K l^'VF^TY

:(• DIVISION

IIIIII lllflllliiiiiii mu

* *-

vaster i'-* > ;

- • - -i

3®fr

I/C

■■- V.

*rf '«

»■? SjM*

'■>** '<■

w I

*. -

: -

«»r

> «

eg ‘ ’sflj”

e?:v

£:

K-

•w»«u.

k"\

tse

•*.£;>• :

« J %w>

- V

ISM-

;. .

5»

ii*-.

. ^

4 t

■A-

>?>'•.

Wfc

• -t . % '

%»

i %

iBai

K>ivvva.. ; )| vSSrS

L?7QC R *<T>s y #3H
/T.V.: ' &P=V‘ rV-5S

:’r

■iii*. A*vT A
> 2-. 1- < - JIKvUC''

ii- V-tV' ’• * r •* -:-w

r£r *! ; v > ? ■-•

,'WLf ,. 'CSE;

•sC? <?■ v >> ,

SNi :

r- v>

V »■

5 vSVv 2gag

■ • ■

•W

AM 1

a -'_f - >*£•

:;.l.u v aE ■ y,.

• » - £ • . •’ ij* /* .* *• • ^ - ->. •■• -■-■■h,',. ; , • , •“?.

-,:■■• t&gmk

Internet Archive Audio

Featured

Top

Images

Featured

Top

Software

Featured

Top

Texts

Featured

Top

Video

Featured

Top

Mobile Apps

Browser Extensions

Archive-It Subscription

Save Page Now

Full text of "An Introduction To Behavioran Introduction To Behavior Theory Theory"

See other formats